From fc85051d6472469d1a301b43aafcb767da7258f4 Mon Sep 17 00:00:00 2001 From: littletomatodonkey <2120160898@bit.edu.cn> Date: Thu, 17 Dec 2020 15:28:19 +0800 Subject: [PATCH 01/60] add support for cpu infer (#1480) * add support for cpu infer * fix readme --- StyleText/README.md | 11 ++++++----- StyleText/README_ch.md | 5 +++-- StyleText/engine/predictors.py | 1 + StyleText/tools/synth_dataset.py | 8 ++++++++ StyleText/tools/synth_image.py | 6 +++--- 5 files changed, 21 insertions(+), 10 deletions(-) diff --git a/StyleText/README.md b/StyleText/README.md index 648b1267..632f7b6d 100644 --- a/StyleText/README.md +++ b/StyleText/README.md @@ -69,12 +69,13 @@ fusion_generator: 1. You can run `tools/synth_image` and generate the demo image, which is saved in the current folder. ```python -python3 -m tools.synth_image -c configs/config.yml --style_image examples/style_images/2.jpg --text_corpus PaddleOCR --language en +python3 tools/synth_image.py -c configs/config.yml --style_image examples/style_images/2.jpg --text_corpus PaddleOCR --language en ``` * Note 1: The language options is correspond to the corpus. Currently, the tool only supports English, Simplified Chinese and Korean. -* Note 2: Synth-Text is mainly used to generate images for OCR recognition models. +* Note 2: Synth-Text is mainly used to generate images for OCR recognition models. So the height of style images should be around 32 pixels. Images in other sizes may behave poorly. +* Note 3: You can modify `use_gpu` in `configs/config.yml` to determine whether to use GPU for prediction. For example, enter the following image and corpus `PaddleOCR`. @@ -122,7 +123,7 @@ In actual application scenarios, it is often necessary to synthesize pictures in * `corpus_file`: Filepath of the corpus. Corpus file should be a text file which will be split by line-endings（'\n'）. Corpus generator samples one line each time. -Example of corpus file: +Example of corpus file: ``` PaddleOCR 飞桨文字识别 @@ -139,9 +140,9 @@ We provide a general dataset containing Chinese, English and Korean (50,000 imag 2. You can run the following command to start synthesis task: ``` bash - python -m tools.synth_dataset.py -c configs/dataset_config.yml + python3 tools/synth_dataset.py -c configs/dataset_config.yml ``` -We also provide example corpus and images in `examples` folder. +We also provide example corpus and images in `examples` folder.
diff --git a/StyleText/README_ch.md b/StyleText/README_ch.md index 0dd5822b..5b8a3ee0 100644 --- a/StyleText/README_ch.md +++ b/StyleText/README_ch.md @@ -61,11 +61,12 @@ fusion_generator: 输入一张风格图和一段文字语料，运行tools/synth_image，合成单张图片，结果图像保存在当前目录下： ```python -python3 -m tools.synth_image -c configs/config.yml --style_image examples/style_images/2.jpg --text_corpus PaddleOCR --language en +python3 tools/synth_image.py -c configs/config.yml --style_image examples/style_images/2.jpg --text_corpus PaddleOCR --language en ``` * 注1：语言选项和语料相对应，目前该工具只支持英文、简体中文和韩语。 * 注2：Style-Text生成的数据主要应用于OCR识别场景。基于当前PaddleOCR识别模型的设计，我们主要支持高度在32左右的风格图像。 如果输入图像尺寸相差过多，效果可能不佳。 +* 注3：可以通过修改配置文件中的`use_gpu`(true或者false)参数来决定是否使用GPU进行预测。 例如，输入如下图片和语料"PaddleOCR": @@ -127,7 +128,7 @@ python3 -m tools.synth_image -c configs/config.yml --style_image examples/style_ 2. 运行`tools/synth_dataset`合成数据： ``` bash - python -m tools.synth_dataset -c configs/dataset_config.yml + python tools/synth_dataset.py -c configs/dataset_config.yml ``` 我们在examples目录下提供了样例图片和语料。
diff --git a/StyleText/engine/predictors.py b/StyleText/engine/predictors.py index d9f4afe4..a1ba21f1 100644 --- a/StyleText/engine/predictors.py +++ b/StyleText/engine/predictors.py @@ -28,6 +28,7 @@ class StyleTextRecPredictor(object): ], "Generator {} not supported.".format(algorithm) use_gpu = config["Global"]['use_gpu'] check_gpu(use_gpu) + paddle.set_device('gpu' if use_gpu else 'cpu') self.logger = get_logger() self.generator = getattr(style_text_rec, algorithm)(config) self.height = config["Global"]["image_height"] diff --git a/StyleText/tools/synth_dataset.py b/StyleText/tools/synth_dataset.py index 4a0e6d5e..a75f7f39 100644 --- a/StyleText/tools/synth_dataset.py +++ b/StyleText/tools/synth_dataset.py @@ -11,6 +11,14 @@ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. + +import os +import sys + +__dir__ = os.path.dirname(os.path.abspath(__file__)) +sys.path.append(__dir__) +sys.path.append(os.path.abspath(os.path.join(__dir__, '..'))) + from engine.synthesisers import DatasetSynthesiser diff --git a/StyleText/tools/synth_image.py b/StyleText/tools/synth_image.py index 7b4827b8..cbc31186 100644 --- a/StyleText/tools/synth_image.py +++ b/StyleText/tools/synth_image.py @@ -16,13 +16,13 @@ import cv2 import sys import glob -from utils.config import ArgsParser -from engine.synthesisers import ImageSynthesiser - __dir__ = os.path.dirname(os.path.abspath(__file__)) sys.path.append(__dir__) sys.path.append(os.path.abspath(os.path.join(__dir__, '..'))) +from utils.config import ArgsParser +from engine.synthesisers import ImageSynthesiser + def synth_image(): args = ArgsParser().parse_args() From fd92294bfa1056bd53339f3ccd6bf2be3c9e936d Mon Sep 17 00:00:00 2001 From: LDOUBLEV Date: Thu, 17 Dec 2020 20:27:43 +0800 Subject: [PATCH 02/60] fludd 2 paddle --- deploy/cpp_infer/readme_en.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/deploy/cpp_infer/readme_en.md b/deploy/cpp_infer/readme_en.md index 30a6912e..8bd76c04 100644 --- a/deploy/cpp_infer/readme_en.md +++ b/deploy/cpp_infer/readme_en.md @@ -107,10 +107,10 @@ make inference_lib_dist For more compilation parameter options, please refer to the official website of the Paddle C++ inference library:[https://www.paddlepaddle.org.cn/documentation/docs/en/advanced_guide/inference_deployment/inference/build_and_install_lib_en.html](https://www.paddlepaddle.org.cn/documentation/docs/en/advanced_guide/inference_deployment/inference/build_and_install_lib_en.html). -* After the compilation process, you can see the following files in the folder of `build/fluid_inference_install_dir/`. +* After the compilation process, you can see the following files in the folder of `build/paddle_inference_install_dir/`. ``` -build/fluid_inference_install_dir/ +build/paddle_inference_install_dir/ |-- CMakeCache.txt |-- paddle |-- third_party From 65b4cc0cf336efb1e6d1ed9ad2f91bd3fbbd1866 Mon Sep 17 00:00:00 2001 From: LDOUBLEV Date: Thu, 17 Dec 2020 20:58:23 +0800 Subject: [PATCH 03/60] adjust rec_batch_num as 1 --- tools/infer/utility.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tools/infer/utility.py b/tools/infer/utility.py index 4b06b60b..1b5e6b52 100755 --- a/tools/infer/utility.py +++ b/tools/infer/utility.py @@ -62,7 +62,7 @@ def parse_args(): parser.add_argument("--rec_model_dir", type=str) parser.add_argument("--rec_image_shape", type=str, default="3, 32, 320") parser.add_argument("--rec_char_type", type=str, default='ch') - parser.add_argument("--rec_batch_num", type=int, default=6) + parser.add_argument("--rec_batch_num", type=int, default=1) parser.add_argument("--max_text_length", type=int, default=25) parser.add_argument( "--rec_char_dict_path", From 30e2bd4f34dd4b566ac4ad3831e33b3c1fcecd53 Mon Sep 17 00:00:00 2001 From: WenmuZhou Date: Fri, 18 Dec 2020 10:04:50 +0800 Subject: [PATCH 04/60] =?UTF-8?q?=E9=80=82=E9=85=8Dpy3.5?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- paddleocr.py | 4 ++-- setup.py | 2 +- 2 files changed, 3 insertions(+), 3 deletions(-) diff --git a/paddleocr.py b/paddleocr.py index 1d8cd254..3c3c47ab 100644 --- a/paddleocr.py +++ b/paddleocr.py @@ -262,8 +262,8 @@ class PaddleOCR(predict_system.TextSystem): logger.error('rec_algorithm must in {}'.format(SUPPORT_REC_MODEL)) sys.exit(0) - postprocess_params.rec_char_dict_path = Path( - __file__).parent / postprocess_params.rec_char_dict_path + postprocess_params.rec_char_dict_path = str( + Path(__file__).parent / postprocess_params.rec_char_dict_path) # init det_model and rec_model super().__init__(postprocess_params) diff --git a/setup.py b/setup.py index f92074be..58f6de48 100644 --- a/setup.py +++ b/setup.py @@ -32,7 +32,7 @@ setup( package_dir={'paddleocr': ''}, include_package_data=True, entry_points={"console_scripts": ["paddleocr= paddleocr.paddleocr:main"]}, - version='2.0.1', + version='2.0.2', install_requires=requirements, license='Apache License 2.0', description='Awesome OCR toolkits based on PaddlePaddle （8.6M ultra-lightweight pre-trained model, support training and deployment among server, mobile, embeded and IoT devices', From a6fd8f80664c458b11e8acea3cd120d30c8834fc Mon Sep 17 00:00:00 2001 From: LDOUBLEV Date: Fri, 18 Dec 2020 11:11:02 +0800 Subject: [PATCH 05/60] adjust cls_batch_num to 6 for speed-up --- tools/infer/utility.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tools/infer/utility.py b/tools/infer/utility.py index 1b5e6b52..162f6cb7 100755 --- a/tools/infer/utility.py +++ b/tools/infer/utility.py @@ -78,7 +78,7 @@ def parse_args(): parser.add_argument("--cls_model_dir", type=str) parser.add_argument("--cls_image_shape", type=str, default="3, 48, 192") parser.add_argument("--label_list", type=list, default=['0', '180']) - parser.add_argument("--cls_batch_num", type=int, default=30) + parser.add_argument("--cls_batch_num", type=int, default=6) parser.add_argument("--cls_thresh", type=float, default=0.9) parser.add_argument("--enable_mkldnn", type=str2bool, default=False) From d489d3c27dbd592a7ad761da6b7e7ee868061499 Mon Sep 17 00:00:00 2001 From: MissPenguin Date: Fri, 18 Dec 2020 03:40:22 +0000 Subject: [PATCH 06/60] fix sast inference bug --- tools/infer/predict_det.py | 3 +++ 1 file changed, 3 insertions(+) diff --git a/tools/infer/predict_det.py b/tools/infer/predict_det.py index ba0adaee..820d3413 100755 --- a/tools/infer/predict_det.py +++ b/tools/infer/predict_det.py @@ -70,6 +70,9 @@ class TextDetector(object): postprocess_params["cover_thresh"] = args.det_east_cover_thresh postprocess_params["nms_thresh"] = args.det_east_nms_thresh elif self.det_algorithm == "SAST": + pre_process_list[0] = { + 'DetResizeForTest': {'resize_long': args.det_limit_side_len} + } postprocess_params['name'] = 'SASTPostProcess' postprocess_params["score_thresh"] = args.det_sast_score_thresh postprocess_params["nms_thresh"] = args.det_sast_nms_thresh From ec37732512e238598c128eef11c9f2f012ce5915 Mon Sep 17 00:00:00 2001 From: LDOUBLEV Date: Fri, 18 Dec 2020 14:45:06 +0800 Subject: [PATCH 07/60] fix db process --- deploy/cpp_infer/src/preprocess_op.cpp | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/deploy/cpp_infer/src/preprocess_op.cpp b/deploy/cpp_infer/src/preprocess_op.cpp index 494b774a..cdb20c31 100644 --- a/deploy/cpp_infer/src/preprocess_op.cpp +++ b/deploy/cpp_infer/src/preprocess_op.cpp @@ -81,14 +81,14 @@ void ResizeImgType0::Run(const cv::Mat &img, cv::Mat &resize_img, else if (resize_h / 32 < 1 + 1e-5) resize_h = 32; else - resize_h = (resize_h / 32 - 1) * 32; + resize_h = (resize_h / 32) * 32; if (resize_w % 32 == 0) resize_w = resize_w; else if (resize_w / 32 < 1 + 1e-5) resize_w = 32; else - resize_w = (resize_w / 32 - 1) * 32; + resize_w = (resize_w / 32) * 32; cv::resize(img, resize_img, cv::Size(resize_w, resize_h)); From 9039cca26d3fcf35fa432e39b299da50df6342b3 Mon Sep 17 00:00:00 2001 From: LDOUBLEV Date: Fri, 18 Dec 2020 15:27:44 +0800 Subject: [PATCH 08/60] add tensorrt args --- tools/infer/predict_det.py | 1 + tools/infer/utility.py | 9 ++++++++- 2 files changed, 9 insertions(+), 1 deletion(-) diff --git a/tools/infer/predict_det.py b/tools/infer/predict_det.py index ba0adaee..1720369d 100755 --- a/tools/infer/predict_det.py +++ b/tools/infer/predict_det.py @@ -35,6 +35,7 @@ logger = get_logger() class TextDetector(object): def __init__(self, args): + self.args = args self.det_algorithm = args.det_algorithm self.use_zero_copy_run = args.use_zero_copy_run pre_process_list = [{ diff --git a/tools/infer/utility.py b/tools/infer/utility.py index 162f6cb7..b793254d 100755 --- a/tools/infer/utility.py +++ b/tools/infer/utility.py @@ -33,6 +33,8 @@ def parse_args(): parser.add_argument("--use_gpu", type=str2bool, default=True) parser.add_argument("--ir_optim", type=str2bool, default=True) parser.add_argument("--use_tensorrt", type=str2bool, default=False) + parser.add_argument("--use_fp16", type=str2bool, default=False) + parser.add_argument("--max_batch_size", type=int, default=10) parser.add_argument("--gpu_mem", type=int, default=8000) # params for text detector @@ -46,7 +48,7 @@ def parse_args(): parser.add_argument("--det_db_thresh", type=float, default=0.3) parser.add_argument("--det_db_box_thresh", type=float, default=0.5) parser.add_argument("--det_db_unclip_ratio", type=float, default=1.6) - + parser.add_argument("--max_batch_size", type=int, default=10) # EAST parmas parser.add_argument("--det_east_score_thresh", type=float, default=0.8) parser.add_argument("--det_east_cover_thresh", type=float, default=0.1) @@ -113,6 +115,11 @@ def create_predictor(args, mode, logger): if args.use_gpu: config.enable_use_gpu(args.gpu_mem, 0) + if args.use_tensorrt: + config.enable_tensorrt_engine( + precision_mode=AnalysisConfig.Precision.Half + if args.use_fp16 else AnalysisConfig.Precision.Float32, + max_batch_size=args.max_batch_size) else: config.disable_gpu() config.set_cpu_math_library_num_threads(6) From e3a2f818fa95563df7b81baf8eacafb5f2da0460 Mon Sep 17 00:00:00 2001 From: WenmuZhou Date: Fri, 18 Dec 2020 17:46:48 +0800 Subject: [PATCH 09/60] fix typo error --- deploy/cpp_infer/tools/config.txt | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/deploy/cpp_infer/tools/config.txt b/deploy/cpp_infer/tools/config.txt index 95d7989b..f1ab0b11 100644 --- a/deploy/cpp_infer/tools/config.txt +++ b/deploy/cpp_infer/tools/config.txt @@ -11,7 +11,7 @@ max_side_len 960 det_db_thresh 0.3 det_db_box_thresh 0.5 det_db_unclip_ratio 2.0 -det_model_dir ./inference/ch__ppocr_mobile_v2.0_det_infer/ +det_model_dir ./inference/ch_ppocr_mobile_v2.0_det_infer/ # cls config use_angle_cls 0 From 7f76986c288d7a340c21012a32faf72f80053e96 Mon Sep 17 00:00:00 2001 From: WenmuZhou Date: Fri, 18 Dec 2020 18:29:13 +0800 Subject: [PATCH 10/60] Remove duplicate definitions --- tools/infer/utility.py | 1 - 1 file changed, 1 deletion(-) diff --git a/tools/infer/utility.py b/tools/infer/utility.py index b793254d..c3d294e6 100755 --- a/tools/infer/utility.py +++ b/tools/infer/utility.py @@ -34,7 +34,6 @@ def parse_args(): parser.add_argument("--ir_optim", type=str2bool, default=True) parser.add_argument("--use_tensorrt", type=str2bool, default=False) parser.add_argument("--use_fp16", type=str2bool, default=False) - parser.add_argument("--max_batch_size", type=int, default=10) parser.add_argument("--gpu_mem", type=int, default=8000) # params for text detector From ae124590150341ca11a833f5f74f175c7fb7799a Mon Sep 17 00:00:00 2001 From: WenmuZhou Date: Fri, 18 Dec 2020 18:51:19 +0800 Subject: [PATCH 11/60] Save configuration files and logs only during training --- tools/program.py | 22 ++++++++++++---------- tools/train.py | 2 +- 2 files changed, 13 insertions(+), 11 deletions(-) diff --git a/tools/program.py b/tools/program.py index 787a59d4..4331f9d4 100755 --- a/tools/program.py +++ b/tools/program.py @@ -332,7 +332,7 @@ def eval(model, valid_dataloader, post_process_class, eval_class): return metirc -def preprocess(): +def preprocess(is_train=False): FLAGS = ArgsParser().parse_args() config = load_config(FLAGS.config) merge_config(FLAGS.opt) @@ -350,15 +350,17 @@ def preprocess(): device = paddle.set_device(device) config['Global']['distributed'] = dist.get_world_size() != 1 - - # save_config - save_model_dir = config['Global']['save_model_dir'] - os.makedirs(save_model_dir, exist_ok=True) - with open(os.path.join(save_model_dir, 'config.yml'), 'w') as f: - yaml.dump(dict(config), f, default_flow_style=False, sort_keys=False) - - logger = get_logger( - name='root', log_file='{}/train.log'.format(save_model_dir)) + if is_train: + # save_config + save_model_dir = config['Global']['save_model_dir'] + os.makedirs(save_model_dir, exist_ok=True) + with open(os.path.join(save_model_dir, 'config.yml'), 'w') as f: + yaml.dump( + dict(config), f, default_flow_style=False, sort_keys=False) + log_file = '{}/train.log'.format(save_model_dir) + else: + log_file = None + logger = get_logger(name='root', log_file=log_file) if config['Global']['use_visualdl']: from visualdl import LogWriter vdl_writer_path = '{}/vdl/'.format(save_model_dir) diff --git a/tools/train.py b/tools/train.py index 6e44c598..383f8d83 100755 --- a/tools/train.py +++ b/tools/train.py @@ -110,6 +110,6 @@ def test_reader(config, device, logger): if __name__ == '__main__': - config, device, logger, vdl_writer = program.preprocess() + config, device, logger, vdl_writer = program.preprocess(is_train=True) main(config, device, logger, vdl_writer) # test_reader(config, device, logger) From 73d4b41ab03171c0b2782edc3a1e0bf029e45e4a Mon Sep 17 00:00:00 2001 From: WenmuZhou Date: Fri, 18 Dec 2020 22:36:04 +0800 Subject: [PATCH 12/60] update inference doc --- doc/doc_ch/inference.md | 11 ++++++++--- doc/doc_en/inference_en.md | 16 ++++++++++++---- 2 files changed, 20 insertions(+), 7 deletions(-) diff --git a/doc/doc_ch/inference.md b/doc/doc_ch/inference.md index 09303e93..c69c127a 100755 --- a/doc/doc_ch/inference.md +++ b/doc/doc_ch/inference.md @@ -245,7 +245,10 @@ python3 tools/infer/predict_det.py --det_algorithm="SAST" --image_dir="./doc/img 超轻量中文识别模型推理，可以执行如下命令： ``` -python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/ch/word_4.jpg" --rec_model_dir="./inference/rec_crnn/" +# 下载超轻量中文识别模型： +wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar +tar xf ch_ppocr_mobile_v2.0_rec_infer.tar +python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/ch/word_4.jpg" --rec_model_dir="ch_ppocr_mobile_v2.0_rec_infer" ``` ![](../imgs_words/ch/word_4.jpg) @@ -266,7 +269,6 @@ Predicts of ./doc/imgs_words/ch/word_4.jpg:('实力活力', 0.98458153) ``` python3 tools/export_model.py -c configs/rec/rec_r34_vd_none_bilstm_ctc.yml -o Global.pretrained_model=./rec_r34_vd_none_bilstm_ctc_v2.0_train/best_accuracy Global.load_static_weights=False Global.save_inference_dir=./inference/rec_crnn - ``` CRNN 文本识别模型推理，可以执行如下命令： @@ -327,7 +329,10 @@ Predicts of ./doc/imgs_words/korean/1.jpg:('바탕으로', 0.9948904) 方向分类模型推理，可以执行如下命令： ``` -python3 tools/infer/predict_cls.py --image_dir="./doc/imgs_words/ch/word_4.jpg" --cls_model_dir="./inference/cls/" +# 下载超轻量中文方向分类器模型： +wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar +tar xf ch_ppocr_mobile_v2.0_cls_infer.tar +python3 tools/infer/predict_cls.py --image_dir="./doc/imgs_words/ch/word_4.jpg" --cls_model_dir="ch_ppocr_mobile_v2.0_cls_infer" ``` ![](../imgs_words/ch/word_1.jpg) diff --git a/doc/doc_en/inference_en.md b/doc/doc_en/inference_en.md index 3fcd36c0..8742b7ce 100755 --- a/doc/doc_en/inference_en.md +++ b/doc/doc_en/inference_en.md @@ -255,15 +255,18 @@ The following will introduce the lightweight Chinese recognition model inference For lightweight Chinese recognition model inference, you can execute the following commands: ``` -python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/ch/word_4.jpg" --rec_model_dir="./inference/rec_crnn/" +# download CRNN text recognition inference model +wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar +tar xf ch_ppocr_mobile_v2.0_rec_infer.tar +python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words_en/word_10.png" --rec_model_dir="ch_ppocr_mobile_v2.0_rec_infer" ``` -![](../imgs_words/ch/word_4.jpg) +![](../imgs_words_en/word_10.png) After executing the command, the prediction results (recognized text and score) of the above image will be printed on the screen. ```bash -Predicts of ./doc/imgs_words/ch/word_4.jpg:('实力活力', 0.98458153) +Predicts of ./doc/imgs_words_en/word_10.png:('PAIN', 0.9897658) ``` @@ -339,7 +342,12 @@ For angle classification model inference, you can execute the following commands ``` python3 tools/infer/predict_cls.py --image_dir="./doc/imgs_words_en/word_10.png" --cls_model_dir="./inference/cls/" ``` - +``` +# download text angle class inference model： +wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar +tar xf ch_ppocr_mobile_v2.0_cls_infer.tar +python3 tools/infer/predict_cls.py --image_dir="./doc/imgs_words_en/word_10.png" --cls_model_dir="ch_ppocr_mobile_v2.0_cls_infer" +``` ![](../imgs_words_en/word_10.png) After executing the command, the prediction results (classification angle and score) of the above image will be printed on the screen. From b615f6704d9b7e7e58c90fc246ec002d7485adac Mon Sep 17 00:00:00 2001 From: WenmuZhou Date: Fri, 18 Dec 2020 23:00:16 +0800 Subject: [PATCH 13/60] load model by pretrained_model --- doc/doc_ch/angle_class.md | 2 +- doc/doc_ch/detection.md | 6 +++--- doc/doc_ch/recognition.md | 5 ++--- doc/doc_en/angle_class_en.md | 2 +- doc/doc_en/detection_en.md | 6 +++--- doc/doc_en/recognition_en.md | 8 ++++---- 6 files changed, 14 insertions(+), 15 deletions(-) diff --git a/doc/doc_ch/angle_class.md b/doc/doc_ch/angle_class.md index 3f2027b9..846be15f 100644 --- a/doc/doc_ch/angle_class.md +++ b/doc/doc_ch/angle_class.md @@ -117,7 +117,7 @@ python3 tools/eval.py -c configs/cls/cls_mv3.yml -o Global.checkpoints={path/to/ ``` # 预测分类结果 -python3 tools/infer_cls.py -c configs/cls/cls_mv3.yml -o Global.checkpoints={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/ch/word_1.jpg +python3 tools/infer_cls.py -c configs/cls/cls_mv3.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.load_static_weights=false Global.infer_img=doc/imgs_words/ch/word_1.jpg ``` 预测图片： diff --git a/doc/doc_ch/detection.md b/doc/doc_ch/detection.md index 08b94a9c..8f0f6979 100644 --- a/doc/doc_ch/detection.md +++ b/doc/doc_ch/detection.md @@ -120,16 +120,16 @@ python3 tools/eval.py -c configs/det/det_mv3_db.yml -o Global.checkpoints="{pat 测试单张图像的检测效果 ```shell -python3 tools/infer_det.py -c configs/det/det_mv3_db.yml -o Global.infer_img="./doc/imgs_en/img_10.jpg" Global.checkpoints="./output/det_db/best_accuracy" +python3 tools/infer_det.py -c configs/det/det_mv3_db.yml -o Global.infer_img="./doc/imgs_en/img_10.jpg" Global.pretrained_model="./output/det_db/best_accuracy" Global.load_static_weights=false ``` 测试DB模型时，调整后处理阈值， ```shell -python3 tools/infer_det.py -c configs/det/det_mv3_db.yml -o Global.infer_img="./doc/imgs_en/img_10.jpg" Global.checkpoints="./output/det_db/best_accuracy" PostProcess.box_thresh=0.6 PostProcess.unclip_ratio=1.5 +python3 tools/infer_det.py -c configs/det/det_mv3_db.yml -o Global.infer_img="./doc/imgs_en/img_10.jpg" Global.pretrained_model="./output/det_db/best_accuracy" Global.load_static_weights=false PostProcess.box_thresh=0.6 PostProcess.unclip_ratio=1.5 ``` 测试文件夹下所有图像的检测效果 ```shell -python3 tools/infer_det.py -c configs/det/det_mv3_db.yml -o Global.infer_img="./doc/imgs_en/" Global.checkpoints="./output/det_db/best_accuracy" +python3 tools/infer_det.py -c configs/det/det_mv3_db.yml -o Global.infer_img="./doc/imgs_en/" Global.pretrained_model="./output/det_db/best_accuracy" Global.load_static_weights=false ``` diff --git a/doc/doc_ch/recognition.md b/doc/doc_ch/recognition.md index dc06365c..b473f3ac 100644 --- a/doc/doc_ch/recognition.md +++ b/doc/doc_ch/recognition.md @@ -324,7 +324,6 @@ Eval: 评估数据集可以通过 `configs/rec/rec_icdar15_train.yml` 修改Eval中的 `label_file_path` 设置。 -*注意* 评估时必须确保配置文件中 infer_img 字段为空 ``` # GPU 评估， Global.checkpoints 为待测权重 python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_icdar15_train.yml -o Global.checkpoints={path/to/weights}/best_accuracy @@ -342,7 +341,7 @@ python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec ``` # 预测英文结果 -python3 tools/infer_rec.py -c configs/rec/rec_icdar15_train.yml -o Global.checkpoints={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png +python3 tools/infer_rec.py -c configs/rec/rec_icdar15_train.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.load_static_weights=false Global.infer_img=doc/imgs_words/en/word_1.png ``` 预测图片： @@ -361,7 +360,7 @@ infer_img: doc/imgs_words/en/word_1.png ``` # 预测中文结果 -python3 tools/infer_rec.py -c configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml -o Global.checkpoints={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/ch/word_1.jpg +python3 tools/infer_rec.py -c configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.load_static_weights=false Global.infer_img=doc/imgs_words/ch/word_1.jpg ``` 预测图片： diff --git a/doc/doc_en/angle_class_en.md b/doc/doc_en/angle_class_en.md index 4c479e7b..e6157d16 100644 --- a/doc/doc_en/angle_class_en.md +++ b/doc/doc_en/angle_class_en.md @@ -119,7 +119,7 @@ Use `Global.infer_img` to specify the path of the predicted picture or folder, a ``` # Predict English results -python3 tools/infer_cls.py -c configs/cls/cls_mv3.yml -o Global.checkpoints={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words_en/word_10.png +python3 tools/infer_cls.py -c configs/cls/cls_mv3.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.load_static_weights=false Global.infer_img=doc/imgs_words_en/word_10.png ``` Input image: diff --git a/doc/doc_en/detection_en.md b/doc/doc_en/detection_en.md index 7638315a..5c4a63e2 100644 --- a/doc/doc_en/detection_en.md +++ b/doc/doc_en/detection_en.md @@ -113,16 +113,16 @@ python3 tools/eval.py -c configs/det/det_mv3_db.yml -o Global.checkpoints="{pat Test the detection result on a single image: ```shell -python3 tools/infer_det.py -c configs/det/det_mv3_db.yml -o Global.infer_img="./doc/imgs_en/img_10.jpg" Global.checkpoints="./output/det_db/best_accuracy" +python3 tools/infer_det.py -c configs/det/det_mv3_db.yml -o Global.infer_img="./doc/imgs_en/img_10.jpg" Global.pretrained_model="./output/det_db/best_accuracy" Global.load_static_weights=false ``` When testing the DB model, adjust the post-processing threshold: ```shell -python3 tools/infer_det.py -c configs/det/det_mv3_db.yml -o Global.infer_img="./doc/imgs_en/img_10.jpg" Global.checkpoints="./output/det_db/best_accuracy" PostProcess.box_thresh=0.6 PostProcess.unclip_ratio=1.5 +python3 tools/infer_det.py -c configs/det/det_mv3_db.yml -o Global.infer_img="./doc/imgs_en/img_10.jpg" Global.pretrained_model="./output/det_db/best_accuracy" Global.load_static_weights=false PostProcess.box_thresh=0.6 PostProcess.unclip_ratio=1.5 ``` Test the detection result on all images in the folder: ```shell -python3 tools/infer_det.py -c configs/det/det_mv3_db.yml -o Global.infer_img="./doc/imgs_en/" Global.checkpoints="./output/det_db/best_accuracy" +python3 tools/infer_det.py -c configs/det/det_mv3_db.yml -o Global.infer_img="./doc/imgs_en/" Global.pretrained_model="./output/det_db/best_accuracy" Global.load_static_weights=false ``` diff --git a/doc/doc_en/recognition_en.md b/doc/doc_en/recognition_en.md index bc8faa0f..7723d20b 100644 --- a/doc/doc_en/recognition_en.md +++ b/doc/doc_en/recognition_en.md @@ -317,11 +317,11 @@ Eval: ### EVALUATION -The evaluation data set can be modified via `configs/rec/rec_icdar15_reader.yml` setting of `label_file_path` in EvalReader. +The evaluation dataset can be set by modifying the `Eval.dataset.label_file_list` field in the `configs/rec/rec_icdar15_train.yml` file. ``` # GPU evaluation, Global.checkpoints is the weight to be tested -python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_icdar15_reader.yml -o Global.checkpoints={path/to/weights}/best_accuracy +python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_icdar15_train.yml -o Global.checkpoints={path/to/weights}/best_accuracy ``` @@ -336,7 +336,7 @@ The default prediction picture is stored in `infer_img`, and the weight is speci ``` # Predict English results -python3 tools/infer_rec.py -c configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml -o Global.checkpoints={path/to/weights}/best_accuracy TestReader.infer_img=doc/imgs_words/en/word_1.jpg +python3 tools/infer_rec.py -c configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.load_static_weights=false Global.infer_img=doc/imgs_words/en/word_1.jpg ``` Input image: @@ -354,7 +354,7 @@ The configuration file used for prediction must be consistent with the training. ``` # Predict Chinese results -python3 tools/infer_rec.py -c configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml -o Global.checkpoints={path/to/weights}/best_accuracy TestReader.infer_img=doc/imgs_words/ch/word_1.jpg +python3 tools/infer_rec.py -c configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.load_static_weights=false Global.infer_img=doc/imgs_words/ch/word_1.jpg ``` Input image: From fd3ddebbbfd4c6c8557bf4e9c1fd4d3b23a61aad Mon Sep 17 00:00:00 2001 From: dyning Date: Sat, 19 Dec 2020 20:02:39 +0800 Subject: [PATCH 14/60] Update README_ch.md --- StyleText/README_ch.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/StyleText/README_ch.md b/StyleText/README_ch.md index 5b8a3ee0..68dc5a21 100644 --- a/StyleText/README_ch.md +++ b/StyleText/README_ch.md @@ -21,7 +21,7 @@
-Style-Text数据合成工具是基于百度自研的文本编辑算法《Editing Text in the Wild》https://arxiv.org/abs/1908.03047 +Style-Text数据合成工具是基于文本编辑算法《Editing Text in the Wild》https://arxiv.org/abs/1908.03047 不同于常用的基于GAN的数据合成工具，Style-Text主要框架包括：1.文本前景风格迁移模块 2.背景抽取模块 3.融合模块。经过这样三步，就可以迅速实现图像文本风格迁移。下图是一些该数据合成工具效果图。 From fa1d09905510e67bd15799ea225519610faedfba Mon Sep 17 00:00:00 2001 From: dyning Date: Sat, 19 Dec 2020 20:02:58 +0800 Subject: [PATCH 15/60] Update README.md --- StyleText/README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/StyleText/README.md b/StyleText/README.md index 60a9ee99..d23cf5db 100644 --- a/StyleText/README.md +++ b/StyleText/README.md @@ -22,7 +22,7 @@ English | [简体中文](README_ch.md)
-The Style-Text data synthesis tool is a tool based on Baidu's self-developed text editing algorithm "Editing Text in the Wild" [https://arxiv.org/abs/1908.03047](https://arxiv.org/abs/1908.03047). +The Style-Text data synthesis tool is a tool based on text editing algorithm "Editing Text in the Wild" [https://arxiv.org/abs/1908.03047](https://arxiv.org/abs/1908.03047). Different from the commonly used GAN-based data synthesis tools, the main framework of Style-Text includes: * (1) Text foreground style transfer module. From 20a7688d23bf4c32f7c0303e56adef5f31b99370 Mon Sep 17 00:00:00 2001 From: dyning Date: Sat, 19 Dec 2020 20:07:06 +0800 Subject: [PATCH 16/60] Update README_ch.md --- StyleText/README_ch.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/StyleText/README_ch.md b/StyleText/README_ch.md index 68dc5a21..ca7f7122 100644 --- a/StyleText/README_ch.md +++ b/StyleText/README_ch.md @@ -21,7 +21,7 @@ -Style-Text数据合成工具是基于文本编辑算法《Editing Text in the Wild》https://arxiv.org/abs/1908.03047 +Style-Text数据合成工具是基于百度和华科合作研发的文本编辑算法《Editing Text in the Wild》https://arxiv.org/abs/1908.03047 不同于常用的基于GAN的数据合成工具，Style-Text主要框架包括：1.文本前景风格迁移模块 2.背景抽取模块 3.融合模块。经过这样三步，就可以迅速实现图像文本风格迁移。下图是一些该数据合成工具效果图。 From 0bc91f49470b0dcae363b140955f119ea47abc47 Mon Sep 17 00:00:00 2001 From: dyning Date: Sat, 19 Dec 2020 20:10:54 +0800 Subject: [PATCH 17/60] Update README.md --- StyleText/README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/StyleText/README.md b/StyleText/README.md index d23cf5db..a015bafb 100644 --- a/StyleText/README.md +++ b/StyleText/README.md @@ -22,7 +22,7 @@ English | [简体中文](README_ch.md) -The Style-Text data synthesis tool is a tool based on text editing algorithm "Editing Text in the Wild" [https://arxiv.org/abs/1908.03047](https://arxiv.org/abs/1908.03047). +The Style-Text data synthesis tool is a tool based on Baidu and HUST cooperation research work, text editing algorithm "Editing Text in the Wild" [https://arxiv.org/abs/1908.03047](https://arxiv.org/abs/1908.03047). Different from the commonly used GAN-based data synthesis tools, the main framework of Style-Text includes: * (1) Text foreground style transfer module. From 9d1bc3ebb070282d691d1730ed01a122aedd34d1 Mon Sep 17 00:00:00 2001 From: dyning Date: Sat, 19 Dec 2020 20:11:21 +0800 Subject: [PATCH 18/60] Update README.md --- StyleText/README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/StyleText/README.md b/StyleText/README.md index a015bafb..697f439c 100644 --- a/StyleText/README.md +++ b/StyleText/README.md @@ -22,7 +22,7 @@ English | [简体中文](README_ch.md) -The Style-Text data synthesis tool is a tool based on Baidu and HUST cooperation research work, text editing algorithm "Editing Text in the Wild" [https://arxiv.org/abs/1908.03047](https://arxiv.org/abs/1908.03047). +The Style-Text data synthesis tool is a tool based on Baidu and HUST cooperation research work, "Editing Text in the Wild" [https://arxiv.org/abs/1908.03047](https://arxiv.org/abs/1908.03047). Different from the commonly used GAN-based data synthesis tools, the main framework of Style-Text includes: * (1) Text foreground style transfer module. From 87eca8ff9c77804ac40dd2c7354edbbba51163d1 Mon Sep 17 00:00:00 2001 From: dyning Date: Sun, 20 Dec 2020 08:20:31 +0800 Subject: [PATCH 19/60] Update README_ch.md --- README_ch.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README_ch.md b/README_ch.md index afdd0536..b1cb778b 100755 --- a/README_ch.md +++ b/README_ch.md @@ -115,7 +115,7 @@ PaddleOCR同时支持动态图与静态图两种编程范式 -PP-OCR是一个实用的超轻量OCR系统。主要由DB文本检测、检测框矫正和CRNN文本识别三部分组成。该系统从骨干网络选择和调整、预测头部的设计、数据增强、学习率变换策略、正则化参数选择、预训练模型使用以及模型自动裁剪量化8个方面，采用19个有效策略，对各个模块的模型进行效果调优和瘦身，最终得到整体大小为3.5M的超轻量中英文OCR和2.8M的英文数字OCR。更多细节请参考PP-OCR技术方案 https://arxiv.org/abs/2009.09941 。其中FPGM裁剪器和PACT量化的实现可以参考[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim)。 +PP-OCR是一个实用的超轻量OCR系统。主要由DB文本检测[2]、检测框矫正和CRNN文本识别三部分组成[7]。该系统从骨干网络选择和调整、预测头部的设计、数据增强、学习率变换策略、正则化参数选择、预训练模型使用以及模型自动裁剪量化8个方面，采用19个有效策略，对各个模块的模型进行效果调优和瘦身，最终得到整体大小为3.5M的超轻量中英文OCR和2.8M的英文数字OCR。更多细节请参考PP-OCR技术方案 https://arxiv.org/abs/2009.09941 。其中FPGM裁剪器[8]和PACT量化[9]的实现可以参考[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim)。 ## 效果展示 [more](./doc/doc_ch/visualization.md) From 75a281f4dbe98a325a2c60b96a8305a5362f60d7 Mon Sep 17 00:00:00 2001 From: dyning Date: Sun, 20 Dec 2020 08:20:45 +0800 Subject: [PATCH 20/60] Update README.md --- README.md | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/README.md b/README.md index 3f6737f8..9bb055cd 100644 --- a/README.md +++ b/README.md @@ -122,8 +122,7 @@ For a new language request, please refer to [Guideline for new language_requests -PP-OCR is a practical ultra-lightweight OCR system. It is mainly composed of three parts: DB text detection, detection frame correction and CRNN text recognition. The system adopts 19 effective strategies from 8 aspects including backbone network selection and adjustment, prediction head design, data augmentation, learning rate transformation strategy, regularization parameter selection, pre-training model use, and automatic model tailoring and quantization to optimize and slim down the models of each module. The final results are an ultra-lightweight Chinese and English OCR model with an overall size of 3.5M and a 2.8M English digital OCR model. For more details, please refer to the PP-OCR technical article (https://arxiv.org/abs/2009.09941). Besides, The implementation of the FPGM Pruner and PACT quantization is based on [PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim). - +PP-OCR is a practical ultra-lightweight OCR system. It is mainly composed of three parts: DB text detection[2], detection frame correction and CRNN text recognition[7]. The system adopts 19 effective strategies from 8 aspects including backbone network selection and adjustment, prediction head design, data augmentation, learning rate transformation strategy, regularization parameter selection, pre-training model use, and automatic model tailoring and quantization to optimize and slim down the models of each module. The final results are an ultra-lightweight Chinese and English OCR model with an overall size of 3.5M and a 2.8M English digital OCR model. For more details, please refer to the PP-OCR technical article (https://arxiv.org/abs/2009.09941). Besides, The implementation of the FPGM Pruner [8] and PACT quantization [9] is based on [PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim). ## Visualization [more](./doc/doc_en/visualization_en.md) From 623843881bad6534edd57cf64b40b9064baca420 Mon Sep 17 00:00:00 2001 From: dyning Date: Sun, 20 Dec 2020 08:25:58 +0800 Subject: [PATCH 22/60] Update algorithm_overview_en.md --- doc/doc_en/algorithm_overview_en.md | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) diff --git a/doc/doc_en/algorithm_overview_en.md b/doc/doc_en/algorithm_overview_en.md index 7f1afd02..b0ef5022 100755 --- a/doc/doc_en/algorithm_overview_en.md +++ b/doc/doc_en/algorithm_overview_en.md @@ -11,9 +11,9 @@ This tutorial lists the text detection algorithms and text recognition algorithm ### 1. Text Detection Algorithm PaddleOCR open source text detection algorithms list: -- [x] EAST([paper](https://arxiv.org/abs/1704.03155)) -- [x] DB([paper](https://arxiv.org/abs/1911.08947)) -- [x] SAST([paper](https://arxiv.org/abs/1908.05498) )(Baidu Self-Research) +- [x] EAST[2]([paper](https://arxiv.org/abs/1704.03155)) +- [x] DB[1]([paper](https://arxiv.org/abs/1911.08947)) +- [x] SAST[4]([paper](https://arxiv.org/abs/1908.05498) )(Baidu Self-Research) On the ICDAR2015 dataset, the text detection result is as follows: @@ -39,11 +39,11 @@ For the training guide and use of PaddleOCR text detection algorithms, please re ### 2. Text Recognition Algorithm PaddleOCR open-source text recognition algorithms list: -- [x] CRNN([paper](https://arxiv.org/abs/1507.05717)) -- [x] Rosetta([paper](https://arxiv.org/abs/1910.05085)) -- [ ] STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html)) coming soon -- [ ] RARE([paper](https://arxiv.org/abs/1603.03915v1)) coming soon -- [ ] SRN([paper](https://arxiv.org/abs/2003.12294) )(Baidu Self-Research) coming soon +- [x] CRNN[7]([paper](https://arxiv.org/abs/1507.05717)) +- [x] Rosetta[10]([paper](https://arxiv.org/abs/1910.05085)) +- [ ] STAR-Net[11]([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html)) coming soon +- [ ] RARE[12]([paper](https://arxiv.org/abs/1603.03915v1)) coming soon +- [ ] SRN[5]([paper](https://arxiv.org/abs/2003.12294) )(Baidu Self-Research) coming soon Refer to [DTRB](https://arxiv.org/abs/1904.01906), the training and evaluation result of these above text recognition (using MJSynth and SynthText for training, evaluate on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE) is as follow: From fc84375dc62e3b7ec68b4e605ee7367861a021a0 Mon Sep 17 00:00:00 2001 From: dyning Date: Sun, 20 Dec 2020 08:26:02 +0800 Subject: [PATCH 23/60] Update algorithm_overview.md --- doc/doc_ch/algorithm_overview.md | 18 +++++++++--------- 1 file changed, 9 insertions(+), 9 deletions(-) diff --git a/doc/doc_ch/algorithm_overview.md b/doc/doc_ch/algorithm_overview.md index a23bfcb1..a7d73042 100755 --- a/doc/doc_ch/algorithm_overview.md +++ b/doc/doc_ch/algorithm_overview.md @@ -9,9 +9,9 @@ ### 1.文本检测算法 PaddleOCR开源的文本检测算法列表： -- [x] DB([paper]( https://arxiv.org/abs/1911.08947) )（ppocr推荐） -- [x] EAST([paper](https://arxiv.org/abs/1704.03155)) -- [x] SAST([paper](https://arxiv.org/abs/1908.05498)) +- [x] DB[2]([paper]( https://arxiv.org/abs/1911.08947) )（ppocr推荐） +- [x] EAST[1]([paper](https://arxiv.org/abs/1704.03155)) +- [x] SAST[4]([paper](https://arxiv.org/abs/1908.05498)) 在ICDAR2015文本检测公开数据集上，算法效果如下： @@ -38,13 +38,13 @@ PaddleOCR文本检测算法的训练和使用请参考文档教程中[模型训 ### 2.文本识别算法 PaddleOCR基于动态图开源的文本识别算法列表： -- [x] CRNN([paper](https://arxiv.org/abs/1507.05717) )（ppocr推荐） -- [x] Rosetta([paper](https://arxiv.org/abs/1910.05085)) -- [ ] STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html)) coming soon -- [ ] RARE([paper](https://arxiv.org/abs/1603.03915v1)) coming soon -- [ ] SRN([paper](https://arxiv.org/abs/2003.12294)) coming soon +- [x] CRNN[7]([paper](https://arxiv.org/abs/1507.05717) )（ppocr推荐） +- [x] Rosetta[10]([paper](https://arxiv.org/abs/1910.05085)) +- [ ] STAR-Net[11]([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html)) coming soon +- [ ] RARE[12]([paper](https://arxiv.org/abs/1603.03915v1)) coming soon +- [ ] SRN[5]([paper](https://arxiv.org/abs/2003.12294)) coming soon -参考[DTRB](https://arxiv.org/abs/1904.01906)文字识别训练和评估流程，使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法效果如下： +参考[DTRB][3](https://arxiv.org/abs/1904.01906)文字识别训练和评估流程，使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法效果如下： |模型|骨干网络|Avg Accuracy|模型存储命名|下载链接| |-|-|-|-|-| From 65af573fb0787b1abada32a5baf71020e2ac77aa Mon Sep 17 00:00:00 2001 From: dyning Date: Sun, 20 Dec 2020 08:26:44 +0800 Subject: [PATCH 24/60] Update algorithm_overview.md --- doc/doc_ch/algorithm_overview.md | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/doc/doc_ch/algorithm_overview.md b/doc/doc_ch/algorithm_overview.md index a7d73042..6ea90d2c 100755 --- a/doc/doc_ch/algorithm_overview.md +++ b/doc/doc_ch/algorithm_overview.md @@ -9,9 +9,9 @@ ### 1.文本检测算法 PaddleOCR开源的文本检测算法列表： -- [x] DB[2]([paper]( https://arxiv.org/abs/1911.08947) )（ppocr推荐） -- [x] EAST[1]([paper](https://arxiv.org/abs/1704.03155)) -- [x] SAST[4]([paper](https://arxiv.org/abs/1908.05498)) +- [x] DB([paper]( https://arxiv.org/abs/1911.08947)) [2]（ppocr推荐） +- [x] EAST([paper](https://arxiv.org/abs/1704.03155))[1] +- [x] SAST([paper](https://arxiv.org/abs/1908.05498))[4] 在ICDAR2015文本检测公开数据集上，算法效果如下： From e26a060b994ba80e8ddaf9531c3705d0a5c20c4f Mon Sep 17 00:00:00 2001 From: dyning Date: Sun, 20 Dec 2020 08:27:20 +0800 Subject: [PATCH 25/60] Update algorithm_overview.md --- doc/doc_ch/algorithm_overview.md | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/doc/doc_ch/algorithm_overview.md b/doc/doc_ch/algorithm_overview.md index 6ea90d2c..361e251b 100755 --- a/doc/doc_ch/algorithm_overview.md +++ b/doc/doc_ch/algorithm_overview.md @@ -38,11 +38,11 @@ PaddleOCR文本检测算法的训练和使用请参考文档教程中[模型训 ### 2.文本识别算法 PaddleOCR基于动态图开源的文本识别算法列表： -- [x] CRNN[7]([paper](https://arxiv.org/abs/1507.05717) )（ppocr推荐） -- [x] Rosetta[10]([paper](https://arxiv.org/abs/1910.05085)) -- [ ] STAR-Net[11]([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html)) coming soon -- [ ] RARE[12]([paper](https://arxiv.org/abs/1603.03915v1)) coming soon -- [ ] SRN[5]([paper](https://arxiv.org/abs/2003.12294)) coming soon +- [x] CRNN([paper](https://arxiv.org/abs/1507.05717))[7]（ppocr推荐） +- [x] Rosetta([paper](https://arxiv.org/abs/1910.05085))[10] +- [ ] STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html))[11] coming soon +- [ ] RARE([paper](https://arxiv.org/abs/1603.03915v1))[12] coming soon +- [ ] SRN([paper](https://arxiv.org/abs/2003.12294))[5] coming soon 参考[DTRB][3](https://arxiv.org/abs/1904.01906)文字识别训练和评估流程，使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法效果如下： From e75c3c1033033e7079f646a21a4bbf307882b423 Mon Sep 17 00:00:00 2001 From: dyning Date: Sun, 20 Dec 2020 08:28:52 +0800 Subject: [PATCH 26/60] Update algorithm_overview_en.md --- doc/doc_en/algorithm_overview_en.md | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) diff --git a/doc/doc_en/algorithm_overview_en.md b/doc/doc_en/algorithm_overview_en.md index b0ef5022..427b3de6 100755 --- a/doc/doc_en/algorithm_overview_en.md +++ b/doc/doc_en/algorithm_overview_en.md @@ -11,9 +11,9 @@ This tutorial lists the text detection algorithms and text recognition algorithm ### 1. Text Detection Algorithm PaddleOCR open source text detection algorithms list: -- [x] EAST[2]([paper](https://arxiv.org/abs/1704.03155)) -- [x] DB[1]([paper](https://arxiv.org/abs/1911.08947)) -- [x] SAST[4]([paper](https://arxiv.org/abs/1908.05498) )(Baidu Self-Research) +- [x] EAST([paper](https://arxiv.org/abs/1704.03155))[2] +- [x] DB([paper](https://arxiv.org/abs/1911.08947))[1] +- [x] SAST([paper](https://arxiv.org/abs/1908.05498))[4] On the ICDAR2015 dataset, the text detection result is as follows: @@ -39,11 +39,11 @@ For the training guide and use of PaddleOCR text detection algorithms, please re ### 2. Text Recognition Algorithm PaddleOCR open-source text recognition algorithms list: -- [x] CRNN[7]([paper](https://arxiv.org/abs/1507.05717)) -- [x] Rosetta[10]([paper](https://arxiv.org/abs/1910.05085)) -- [ ] STAR-Net[11]([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html)) coming soon -- [ ] RARE[12]([paper](https://arxiv.org/abs/1603.03915v1)) coming soon -- [ ] SRN[5]([paper](https://arxiv.org/abs/2003.12294) )(Baidu Self-Research) coming soon +- [x] CRNN([paper](https://arxiv.org/abs/1507.05717))[7] +- [x] Rosetta([paper](https://arxiv.org/abs/1910.05085))[10] +- [ ] STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html))[11] coming soon +- [ ] RARE([paper](https://arxiv.org/abs/1603.03915v1))[12] coming soon +- [ ] SRN([paper](https://arxiv.org/abs/2003.12294))[5] coming soon Refer to [DTRB](https://arxiv.org/abs/1904.01906), the training and evaluation result of these above text recognition (using MJSynth and SynthText for training, evaluate on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE) is as follow: From 42fc648c8736cf6a820d5e1ee662811318b57cfc Mon Sep 17 00:00:00 2001 From: dyning Date: Sun, 20 Dec 2020 08:29:12 +0800 Subject: [PATCH 27/60] Update reference.md --- doc/doc_ch/reference.md | 78 ++++++++++++++++++++++++++++++++++++----- 1 file changed, 69 insertions(+), 9 deletions(-) diff --git a/doc/doc_ch/reference.md b/doc/doc_ch/reference.md index 9d9a6785..f1337ded 100644 --- a/doc/doc_ch/reference.md +++ b/doc/doc_ch/reference.md @@ -11,11 +11,12 @@ } 2. DB: -@article{liao2019real, - title={Real-time Scene Text Detection with Differentiable Binarization}, +@inproceedings{liao2020real, + title={Real-Time Scene Text Detection with Differentiable Binarization.}, author={Liao, Minghui and Wan, Zhaoyi and Yao, Cong and Chen, Kai and Bai, Xiang}, - journal={arXiv preprint arXiv:1911.08947}, - year={2019} + booktitle={AAAI}, + pages={11474--11481}, + year={2020} } 3. DTRB: @@ -37,10 +38,11 @@ } 5. SRN: -@article{yu2020towards, - title={Towards Accurate Scene Text Recognition with Semantic Reasoning Networks}, - author={Yu, Deli and Li, Xuan and Zhang, Chengquan and Han, Junyu and Liu, Jingtuo and Ding, Errui}, - journal={arXiv preprint arXiv:2003.12294}, +@inproceedings{yu2020towards, + title={Towards accurate scene text recognition with semantic reasoning networks}, + author={Yu, Deli and Li, Xuan and Zhang, Chengquan and Liu, Tao and Han, Junyu and Liu, Jingtuo and Ding, Errui}, + booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition}, + pages={12113--12122}, year={2020} } @@ -52,4 +54,62 @@ pages={9086--9095}, year={2019} } -``` \ No newline at end of file + +7. CRNN: +@article{shi2016end, + title={An end-to-end trainable neural network for image-based sequence recognition and its application to scene text recognition}, + author={Shi, Baoguang and Bai, Xiang and Yao, Cong}, + journal={IEEE transactions on pattern analysis and machine intelligence}, + volume={39}, + number={11}, + pages={2298--2304}, + year={2016}, + publisher={IEEE} +} + +8. FPGM: +@inproceedings{he2019filter, + title={Filter pruning via geometric median for deep convolutional neural networks acceleration}, + author={He, Yang and Liu, Ping and Wang, Ziwei and Hu, Zhilan and Yang, Yi}, + booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition}, + pages={4340--4349}, + year={2019} +} + +9. PACT: +@article{choi2018pact, + title={Pact: Parameterized clipping activation for quantized neural networks}, + author={Choi, Jungwook and Wang, Zhuo and Venkataramani, Swagath and Chuang, Pierce I-Jen and Srinivasan, Vijayalakshmi and Gopalakrishnan, Kailash}, + journal={arXiv preprint arXiv:1805.06085}, + year={2018} +} + +10.Rosetta +@inproceedings{borisyuk2018rosetta, + title={Rosetta: Large scale system for text detection and recognition in images}, + author={Borisyuk, Fedor and Gordo, Albert and Sivakumar, Viswanath}, + booktitle={Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery \& Data Mining}, + pages={71--79}, + year={2018} +} + +11.STAR-Net +@inproceedings{liu2016star, + title={STAR-Net: A SpaTial Attention Residue Network for Scene Text Recognition.}, + author={Liu, Wei and Chen, Chaofeng and Wong, Kwan-Yee K and Su, Zhizhong and Han, Junyu}, + booktitle={BMVC}, + volume={2}, + pages={7}, + year={2016} +} + +12.RARE +@inproceedings{shi2016robust, + title={Robust scene text recognition with automatic rectification}, + author={Shi, Baoguang and Wang, Xinggang and Lyu, Pengyuan and Yao, Cong and Bai, Xiang}, + booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition}, + pages={4168--4176}, + year={2016} +} + +``` From e6315ddd76cc73dea641b66b8cec8fbf06418911 Mon Sep 17 00:00:00 2001 From: gnoixnil'il Date: Sun, 20 Dec 2020 13:46:40 +0800 Subject: [PATCH 28/60] Update det_basic_loss.py MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit 删除重复赋值[self.main_loss_type] --- ppocr/losses/det_basic_loss.py | 1 - 1 file changed, 1 deletion(-) diff --git a/ppocr/losses/det_basic_loss.py b/ppocr/losses/det_basic_loss.py index ef656e8c..57b3667d 100644 --- a/ppocr/losses/det_basic_loss.py +++ b/ppocr/losses/det_basic_loss.py @@ -45,7 +45,6 @@ class BalanceLoss(nn.Layer): self.balance_loss = balance_loss self.main_loss_type = main_loss_type self.negative_ratio = negative_ratio - self.main_loss_type = main_loss_type self.return_origin = return_origin self.eps = eps From f103ed2791d1c4aacabec6ef61658d8dc754067f Mon Sep 17 00:00:00 2001 From: LDOUBLEV Date: Mon, 21 Dec 2020 11:22:39 +0800 Subject: [PATCH 29/60] add FAQ --- doc/doc_ch/FAQ.md | 104 ++++++++++++++++++++++++++++++++-------------- 1 file changed, 73 insertions(+), 31 deletions(-) diff --git a/doc/doc_ch/FAQ.md b/doc/doc_ch/FAQ.md index 27b3126c..1e8ebf9b 100755 --- a/doc/doc_ch/FAQ.md +++ b/doc/doc_ch/FAQ.md @@ -14,37 +14,48 @@ * [【理论篇】OCR通用30个问题](#OCR通用问题) * [基础知识7题](#基础知识) * [数据集7题](#数据集2) - * [模型训练调优7题](#模型训练调优2) - * [预测部署9题](#预测部署2) + * [模型训练调优18题](#模型训练调优2) * [【实战篇】PaddleOCR实战87个问题](#PaddleOCR实战问题) - * [使用咨询21题](#使用咨询) + * [使用咨询23题](#使用咨询) * [数据集17题](#数据集3) * [模型训练调优25题](#模型训练调优3) - * [预测部署24题](#预测部署3) + * [预测部署25题](#预测部署3) ## 近期更新（2020.12.14） -#### Q3.1.21：PaddleOCR支持动态图吗？ +#### Q2.3.17: StyleText 合成数据效果不好？ +**A**： StyleText模型生成的数据主要用于OCR识别模型的训练。PaddleOCR目前识别模型的输入为32*N，因此当前版本模型主要适用高度为32的数据。建议要合成的数据尺寸设置为32*N。尺寸相差不多的数据也可以生成，尺寸很大或很小的数据效果确实不佳。 -**A**：动态图版本正在紧锣密鼓开发中，将于2020年12月16日发布，敬请关注。 +#### Q2.3.18: PaddleOCR develop分支和dygraph分支的区别？ +**A**：目前PaddleOCR有四个分支，分别是： +- develop：基于Paddle静态图开发的分支，推荐使用paddle1.8 或者2.0版本，该分支具备完善的模型训练、预测、推理部署、量化裁剪等功能。 +- release/1.1：PaddleOCR 发布的第一个稳定版本，基于静态图开发，具备完善的训练、预测、推理部署、量化裁剪等功能。 +- dygraph：基于Paddle动态图开发的分支，目前仍在开发中，未来将作为主要开发分支，运行要求使用Paddle2.0rc1版本，目前仍在开发中。 +- release/2.0-rc1-0：PaddleOCR发布的第二个稳定版本，基于动态图和paddle2.0版本开发，动态图开发的工程更易于调试，目前支，支持模型训练、预测，暂不支持移动端部署。 -#### Q3.3.23：检测模型训练或预测时出现elementwise_add报错 +如果您已经上手过PaddleOCR，并且希望在各种环境上部署PaddleOCR，目前建议使用静态图分支，develop或者release/1.1分支。如果您是初学者，想快速训练，调试PaddleOCR中的算法，建议尝鲜PaddleOCR dygraph分支。 -**A**：设置的输入尺寸必须是32的倍数，否则在网络多次下采样和上采样后，feature map会产生1个像素的diff，从而导致elementwise_add时报shape不匹配的错误。 +**注意**：develop和dygraph分支要求的Paddle版本、本地环境有差别，请注意不同分支环境安装部分的差异。 -#### Q3.3.24: DB检测训练输入尺寸640，可以改大一些吗？ +#### Q3.1.22： ModuleNotFoundError: No module named 'paddle.nn'， +**A**： paddle.nn是Paddle2.0版本特有的功能，请安装大于等于Paddle 2.0.0rc1的版本，安装方式为 +``` +python3 -m pip install paddlepaddle-gpu==2.0.0rc1 -i https://mirror.baidu.com/pypi/simple +``` -**A**: 不建议改大。检测模型训练输入尺寸是预处理中random crop后的尺寸，并非直接将原图进行resize，多数场景下这个尺寸并不小了，改大后可能反而并不合适，而且训练会变慢。另外，代码里可能有的地方参数按照预设输入尺寸适配的，改大后可能有隐藏风险。 +#### Q3.1.23： ImportError: /usr/lib/x86_64_linux-gnu/libstdc++.so.6:version `CXXABI_1.3.11` not found (required by /usr/lib/python3.6/site-package/paddle/fluid/core+avx.so) +**A**：这个问题是glibc版本不足导致的，Paddle2.0rc1版本对gcc版本和glib版本有更高的要求，推荐gcc版本为8.2，glibc版本2.12以上。 +如果您的环境不满足这个要求，或者使用的docker镜像为: +`hub.baidubce.com/paddlepaddle/paddle:latest-gpu-cuda9.0-cudnn7-dev` +`hub.baidubce.com/paddlepaddle/paddle:latest-gpu-cuda9.0-cudnn7-dev`，安装Paddle2.0rc版本可能会出现上述错误， +2.0版本推荐使用新的docker镜像 `paddlepaddle/paddle:latest-dev-cuda10.1-cudnn7-gcc82`。 +或者访问[dockerhub](https://hub.docker.com/r/paddlepaddle/paddle/tags/)获得与您机器适配的镜像。 -#### Q3.3.25: 识别模型训练时，loss能正常下降，但acc一直为0 +#### Q3.4.25 : PaddleOCR模型Python端预测和C++预测结果不一致？ +**A**：正常来说，python端预测和C++预测文本是一致的，如果预测结果差异较大，建议首先排查diff出现在检测模型还是识别模型，或者尝试换其他模型是否有类似的问题。其次，检查python端和C++端数据处理部分是否存在差异，建议保存环境，更新PaddleOCR代码再试下。如果更新代码或者更新代码都没能解决，建议在PaddleOCR群里或者issue中抛出您的问题。 -**A**: 识别模型训练初期acc为0是正常的，多训一段时间指标就上来了。 - -#### Q3.4.24：DB模型能正确推理预测，但换成EAST或SAST模型时报错或结果不正确 - -**A**：使用EAST或SAST模型进行推理预测时，需要在命令中指定参数--det_algorithm="EAST" 或 --det_algorithm="SAST"，使用DB时不用指定是因为该参数默认值是"DB"：https://github.com/PaddlePaddle/PaddleOCR/blob/e7a708e9fdaf413ed7a14da8e4a7b4ac0b211e42/tools/infer/utility.py#L43 ## 【精选】OCR精选10个问题 @@ -238,18 +249,15 @@ （2）调大系统的[l2 dcay值](https://github.com/PaddlePaddle/PaddleOCR/blob/a501603d54ff5513fc4fc760319472e59da25424/configs/rec/ch_ppocr_v1.1/rec_chinese_lite_train_v1.1.yml#L47) - -### 预测部署 - -#### Q2.4.1：请问对于图片中的密集文字，有什么好的处理办法吗？ +#### Q2.3.8：请问对于图片中的密集文字，有什么好的处理办法吗？ **A**：可以先试用预训练模型测试一下，例如DB+CRNN，判断下密集文字图片中是检测还是识别的问题，然后针对性的改善。还有一种是如果图象中密集文字较小，可以尝试增大图像分辨率，对图像进行一定范围内的拉伸，将文字稀疏化，提高识别效果。 -#### Q2.4.2：对于一些在识别时稍微模糊的文本，有没有一些图像增强的方式？ +#### Q2.3.9：对于一些在识别时稍微模糊的文本，有没有一些图像增强的方式？ **A**：在人类肉眼可以识别的前提下，可以考虑图像处理中的均值滤波、中值滤波或者高斯滤波等模糊算子尝试。也可以尝试从数据扩增扰动来强化模型鲁棒性，另外新的思路有对抗性训练和超分SR思路，可以尝试借鉴。但目前业界尚无普遍认可的最优方案，建议优先在数据采集阶段增加一些限制提升图片质量。 -#### Q2.4.3：对于特定文字检测，例如身份证只检测姓名，检测指定区域文字更好，还是检测全部区域再筛选更好？ +#### Q2.3.10：对于特定文字检测，例如身份证只检测姓名，检测指定区域文字更好，还是检测全部区域再筛选更好？ **A**：两个角度来说明一般检测全部区域再筛选更好。 @@ -257,11 +265,11 @@ （2）产品的需求可能是变化的，不排除后续对于模型需求变化的可能性（比如又需要增加一个字段），相比于训练模型，后处理的逻辑会更容易调整。 -#### Q2.4.4：对于小白如何快速入门中文OCR项目实践？ +#### Q2.3.11：对于小白如何快速入门中文OCR项目实践？ **A**：建议可以先了解OCR方向的基础知识，大概了解基础的检测和识别模型算法。然后在Github上可以查看OCR方向相关的repo。目前来看，从内容的完备性来看，PaddleOCR的中英文双语教程文档是有明显优势的，在数据集、模型训练、预测部署文档详实，可以快速入手。而且还有微信用户群答疑，非常适合学习实践。项目地址：[PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR) -#### Q2.4.5：如何识别带空格的英文行文本图像？ +#### Q2.1.12：如何识别带空格的英文行文本图像？ **A**：空格识别可以考虑以下两种方案： @@ -269,22 +277,36 @@ (2)优化文本识别算法。在识别字典里面引入空格字符，然后在识别的训练数据中，如果用空行，进行标注。此外，合成数据时，通过拼接训练数据，生成含有空格的文本。 -#### Q2.4.6：中英文一起识别时也可以加空格字符来训练吗 +#### Q2.3.13：中英文一起识别时也可以加空格字符来训练吗 **A**：中文识别可以加空格当做分隔符训练，具体的效果如何没法给出直接评判，根据实际业务数据训练来判断。 -#### Q2.4.7：低像素文字或者字号比较小的文字有什么超分辨率方法吗 +#### Q2.3.14：低像素文字或者字号比较小的文字有什么超分辨率方法吗 **A**：超分辨率方法分为传统方法和基于深度学习的方法。基于深度学习的方法中，比较经典的有SRCNN，另外CVPR2020也有一篇超分辨率的工作可以参考文章：Unpaired Image Super-Resolution using Pseudo-Supervision，但是没有充分的实践验证过，需要看实际场景下的效果。 -#### Q2.4.8：表格识别有什么好的模型 或者论文推荐么 +#### Q2.3.15：表格识别有什么好的模型 或者论文推荐么 **A**：表格目前学术界比较成熟的解决方案不多 ，可以尝试下分割的论文方案。 -#### Q2.4.9：弯曲文本有试过opencv的TPS进行弯曲校正吗？ +#### Q2.3.16：弯曲文本有试过opencv的TPS进行弯曲校正吗？ **A**：opencv的tps需要标出上下边界对应的点，这个点很难通过传统方法或者深度学习方法获取。PaddleOCR里StarNet网络中的tps模块实现了自动学点，自动校正，可以直接尝试这个。 +#### Q2.3.17: StyleText 合成数据效果不好？ +**A**：StyleText模型生成的数据主要用于OCR识别模型的训练。PaddleOCR目前识别模型的输入为32*N，因此当前版本模型主要适用高度为32的数据。建议要合成的数据尺寸设置为32*N。尺寸相差不多的数据也可以生成，尺寸很大或很小的数据效果确实不佳。 + +#### Q2.3.18: PaddleOCR develop分支和dygraph分支的区别？ +**A** 目前PaddleOCR有四个分支，分别是： + +- develop：基于Paddle静态图开发的分支，推荐使用paddle1.8 或者2.0版本，该分支具备完善的模型训练、预测、推理部署、量化裁剪等功能。 +- release/1.1：PaddleOCR 发布的第一个稳定版本，基于静态图开发，具备完善的训练、预测、推理部署、量化裁剪等功能。 +- dygraph：基于Paddle动态图开发的分支，目前仍在开发中，未来将作为主要开发分支，运行要求使用Paddle2.0rc1版本，目前仍在开发中。 +- release/2.0-rc1-0：PaddleOCR发布的第二个稳定版本，基于动态图和paddle2.0版本开发，动态图开发的工程更易于调试，目前支，支持模型训练、预测，暂不支持移动端部署。 + +如果您已经上手过PaddleOCR，并且希望在各种环境上部署PaddleOCR，目前建议使用静态图分支，develop或者release/1.1分支。如果您是初学者，想快速训练，调试PaddleOCR中的算法，建议尝鲜PaddleOCR dygraph分支。 + +**注意**：develop和dygraph分支要求的Paddle版本、本地环境有差别，请注意不同分支环境安装部分的差异。 @@ -392,6 +414,20 @@ **A**：动态图版本正在紧锣密鼓开发中，将于2020年12月16日发布，敬请关注。 +#### Q3.1.22：ModuleNotFoundError: No module named 'paddle.nn'， +**A**：paddle.nn是Paddle2.0版本特有的功能，请安装大于等于Paddle 2.0.0rc1的版本，安装方式为 +``` +python3 -m pip install paddlepaddle-gpu==2.0.0rc1 -i https://mirror.baidu.com/pypi/simple +``` + +#### Q3.1.23： ImportError: /usr/lib/x86_64_linux-gnu/libstdc++.so.6:version `CXXABI_1.3.11` not found (required by /usr/lib/python3.6/site-package/paddle/fluid/core+avx.so) +**A**：这个问题是glibc版本不足导致的，Paddle2.0rc1版本对gcc版本和glib版本有更高的要求，推荐gcc版本为8.2，glibc版本2.12以上。 +如果您的环境不满足这个要求，或者使用的docker镜像为: +`hub.baidubce.com/paddlepaddle/paddle:latest-gpu-cuda9.0-cudnn7-dev` +`hub.baidubce.com/paddlepaddle/paddle:latest-gpu-cuda9.0-cudnn7-dev`，安装Paddle2.0rc版本可能会出现上述错误，2.0版本推荐使用新的docker镜像 `paddlepaddle/paddle:latest-dev-cuda10.1-cudnn7-gcc82`。 +或者访问[dockerhub](https://hub.docker.com/r/paddlepaddle/paddle/tags/)获得与您机器适配的镜像。 + + ### 数据集 @@ -594,11 +630,11 @@ ps -axu | grep train.py | awk '{print $2}' | xargs kill -9 #### Q3.3.20: 文字检测时怎么模糊的数据增强？ -**A**: 模糊的数据增强需要修改代码进行添加，以DB为例，参考[Normalize](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppocr/data/imaug/operators.py#L60) ,添加模糊的增强就行 +**A**: 模糊的数据增强需要修改代码进行添加，以DB为例，参考[Normalize](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppocr/data/imaug/operators.py#L60) ,添加模糊的增强就行 #### Q3.3.21: 文字检测时怎么更改图片旋转的角度，实现360度任意旋转？ -**A**: 将[这里](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppocr/data/imaug/iaa_augment.py#L64) 的(-10,10) 改为(-180,180)即可 +**A**: 将[这里](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppocr/data/imaug/iaa_augment.py#L64) 的(-10,10) 改为(-180,180)即可 #### Q3.3.22: 训练数据的长宽比过大怎么修改shape @@ -728,4 +764,10 @@ ps -axu | grep train.py | awk '{print $2}' | xargs kill -9 #### Q3.4.24：DB模型能正确推理预测，但换成EAST或SAST模型时报错或结果不正确 -**A**：使用EAST或SAST模型进行推理预测时，需要在命令中指定参数--det_algorithm="EAST" 或 --det_algorithm="SAST"，使用DB时不用指定是因为该参数默认值是"DB"：https://github.com/PaddlePaddle/PaddleOCR/blob/e7a708e9fdaf413ed7a14da8e4a7b4ac0b211e42/tools/infer/utility.py#L43 \ No newline at end of file +**A**：使用EAST或SAST模型进行推理预测时，需要在命令中指定参数--det_algorithm="EAST" 或 --det_algorithm="SAST"，使用DB时不用指定是因为该参数默认值是"DB"：https://github.com/PaddlePaddle/PaddleOCR/blob/e7a708e9fdaf413ed7a14da8e4a7b4ac0b211e42/tools/infer/utility.py#L43 + +#### Q3.4.25 : PaddleOCR模型Python端预测和C++预测结果不一致？ +正常来说，python端预测和C++预测文本是一致的，如果预测结果差异较大， +建议首先排查diff出现在检测模型还是识别模型，或者尝试换其他模型是否有类似的问题。 +其次，检查python端和C++端数据处理部分是否存在差异，建议保存环境，更新PaddleOCR代码再试下。 +如果更新代码或者更新代码都没能解决，建议在PaddleOCR微信群里或者issue中抛出您的问题。 From a8927059c99f07c6c561bb630d7276b7c203a4e5 Mon Sep 17 00:00:00 2001 From: LDOUBLEV Date: Mon, 21 Dec 2020 11:24:27 +0800 Subject: [PATCH 30/60] * to x --- doc/doc_ch/FAQ.md | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/doc/doc_ch/FAQ.md b/doc/doc_ch/FAQ.md index 1e8ebf9b..a893677f 100755 --- a/doc/doc_ch/FAQ.md +++ b/doc/doc_ch/FAQ.md @@ -26,7 +26,8 @@ ## 近期更新（2020.12.14） #### Q2.3.17: StyleText 合成数据效果不好？ -**A**： StyleText模型生成的数据主要用于OCR识别模型的训练。PaddleOCR目前识别模型的输入为32*N，因此当前版本模型主要适用高度为32的数据。建议要合成的数据尺寸设置为32*N。尺寸相差不多的数据也可以生成，尺寸很大或很小的数据效果确实不佳。 +**A**： StyleText模型生成的数据主要用于OCR识别模型的训练。PaddleOCR目前识别模型的输入为32 x N，因此当前版本模型主要适用高度为32的数据。 +建议要合成的数据尺寸设置为32 x N。尺寸相差不多的数据也可以生成，尺寸很大或很小的数据效果确实不佳。 #### Q2.3.18: PaddleOCR develop分支和dygraph分支的区别？ **A**：目前PaddleOCR有四个分支，分别是： From 520670a4e4d1f8500a68a24370831a1109b6be43 Mon Sep 17 00:00:00 2001 From: LDOUBLEV Date: Mon, 21 Dec 2020 11:26:05 +0800 Subject: [PATCH 31/60] * to x --- doc/doc_ch/FAQ.md | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/doc/doc_ch/FAQ.md b/doc/doc_ch/FAQ.md index a893677f..182abc8c 100755 --- a/doc/doc_ch/FAQ.md +++ b/doc/doc_ch/FAQ.md @@ -295,7 +295,8 @@ python3 -m pip install paddlepaddle-gpu==2.0.0rc1 -i https://mirror.baidu.com/py **A**：opencv的tps需要标出上下边界对应的点，这个点很难通过传统方法或者深度学习方法获取。PaddleOCR里StarNet网络中的tps模块实现了自动学点，自动校正，可以直接尝试这个。 #### Q2.3.17: StyleText 合成数据效果不好？ -**A**：StyleText模型生成的数据主要用于OCR识别模型的训练。PaddleOCR目前识别模型的输入为32*N，因此当前版本模型主要适用高度为32的数据。建议要合成的数据尺寸设置为32*N。尺寸相差不多的数据也可以生成，尺寸很大或很小的数据效果确实不佳。 +**A**：StyleText模型生成的数据主要用于OCR识别模型的训练。PaddleOCR目前识别模型的输入为32 x N，因此当前版本模型主要适用高度为32的数据。 +建议要合成的数据尺寸设置为32 x N。尺寸相差不多的数据也可以生成，尺寸很大或很小的数据效果确实不佳。 #### Q2.3.18: PaddleOCR develop分支和dygraph分支的区别？ **A** 目前PaddleOCR有四个分支，分别是： From 037bd74017b35e0aed8d77a6cede1f064b532835 Mon Sep 17 00:00:00 2001 From: LDOUBLEV Date: Mon, 21 Dec 2020 11:35:46 +0800 Subject: [PATCH 32/60] fix comments --- README_ch.md | 2 +- doc/doc_ch/FAQ.md | 12 ++++++------ 2 files changed, 7 insertions(+), 7 deletions(-) diff --git a/README_ch.md b/README_ch.md index afdd0536..c7ddf0fe 100755 --- a/README_ch.md +++ b/README_ch.md @@ -8,10 +8,10 @@ PaddleOCR同时支持动态图与静态图两种编程范式 - 静态图版本：develop分支 **近期更新** +- 2020.12.21 [FAQ](./doc/doc_ch/FAQ.md)新增5个高频问题，总数132个，每周一都会更新，欢迎大家持续关注。 - 2020.12.15 更新数据合成工具[Style-Text](./StyleText/README_ch.md)，可以批量合成大量与目标场景类似的图像，在多个场景验证，效果明显提升。 - 2020.12.14 [FAQ](./doc/doc_ch/FAQ.md)新增5个高频问题，总数127个，每周一都会更新，欢迎大家持续关注。 - 2020.11.25 更新半自动标注工具[PPOCRLabel](./PPOCRLabel/README_ch.md)，辅助开发者高效完成标注任务，输出格式与PP-OCR训练任务完美衔接。 -- 2020.9.22 更新PP-OCR技术文章，https://arxiv.org/abs/2009.09941 - [More](./doc/doc_ch/update.md) diff --git a/doc/doc_ch/FAQ.md b/doc/doc_ch/FAQ.md index 182abc8c..5740bd0e 100755 --- a/doc/doc_ch/FAQ.md +++ b/doc/doc_ch/FAQ.md @@ -9,13 +9,13 @@ ## PaddleOCR常见问题汇总(持续更新) -* [近期更新（2020.12.14）](#近期更新) +* [近期更新（2020.12.21）](#近期更新) * [【精选】OCR精选10个问题](#OCR精选10个问题) -* [【理论篇】OCR通用30个问题](#OCR通用问题) +* [【理论篇】OCR通用32个问题](#OCR通用问题) * [基础知识7题](#基础知识) * [数据集7题](#数据集2) * [模型训练调优18题](#模型训练调优2) -* [【实战篇】PaddleOCR实战87个问题](#PaddleOCR实战问题) +* [【实战篇】PaddleOCR实战90个问题](#PaddleOCR实战问题) * [使用咨询23题](#使用咨询) * [数据集17题](#数据集3) * [模型训练调优25题](#模型训练调优3) @@ -23,7 +23,7 @@ -## 近期更新（2020.12.14） +## 近期更新（2020.12.21） #### Q2.3.17: StyleText 合成数据效果不好？ **A**： StyleText模型生成的数据主要用于OCR识别模型的训练。PaddleOCR目前识别模型的输入为32 x N，因此当前版本模型主要适用高度为32的数据。 @@ -33,7 +33,7 @@ **A**：目前PaddleOCR有四个分支，分别是： - develop：基于Paddle静态图开发的分支，推荐使用paddle1.8 或者2.0版本，该分支具备完善的模型训练、预测、推理部署、量化裁剪等功能。 - release/1.1：PaddleOCR 发布的第一个稳定版本，基于静态图开发，具备完善的训练、预测、推理部署、量化裁剪等功能。 -- dygraph：基于Paddle动态图开发的分支，目前仍在开发中，未来将作为主要开发分支，运行要求使用Paddle2.0rc1版本，目前仍在开发中。 +- dygraph：基于Paddle动态图开发的分支，未来将作为主要开发分支，运行要求使用Paddle2.0rc1版本，目前仍在开发中。 - release/2.0-rc1-0：PaddleOCR发布的第二个稳定版本，基于动态图和paddle2.0版本开发，动态图开发的工程更易于调试，目前支，支持模型训练、预测，暂不支持移动端部署。 如果您已经上手过PaddleOCR，并且希望在各种环境上部署PaddleOCR，目前建议使用静态图分支，develop或者release/1.1分支。如果您是初学者，想快速训练，调试PaddleOCR中的算法，建议尝鲜PaddleOCR dygraph分支。 @@ -270,7 +270,7 @@ python3 -m pip install paddlepaddle-gpu==2.0.0rc1 -i https://mirror.baidu.com/py **A**：建议可以先了解OCR方向的基础知识，大概了解基础的检测和识别模型算法。然后在Github上可以查看OCR方向相关的repo。目前来看，从内容的完备性来看，PaddleOCR的中英文双语教程文档是有明显优势的，在数据集、模型训练、预测部署文档详实，可以快速入手。而且还有微信用户群答疑，非常适合学习实践。项目地址：[PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR) -#### Q2.1.12：如何识别带空格的英文行文本图像？ +#### Q3.12：如何识别带空格的英文行文本图像？ **A**：空格识别可以考虑以下两种方案： From b2768e4a552f65c059ab59a755aedf5864e4160d Mon Sep 17 00:00:00 2001 From: LDOUBLEV Date: Mon, 21 Dec 2020 11:42:15 +0800 Subject: [PATCH 33/60] fix comments --- doc/doc_ch/FAQ.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/doc/doc_ch/FAQ.md b/doc/doc_ch/FAQ.md index 5740bd0e..275d9ac3 100755 --- a/doc/doc_ch/FAQ.md +++ b/doc/doc_ch/FAQ.md @@ -31,7 +31,7 @@ #### Q2.3.18: PaddleOCR develop分支和dygraph分支的区别？ **A**：目前PaddleOCR有四个分支，分别是： -- develop：基于Paddle静态图开发的分支，推荐使用paddle1.8 或者2.0版本，该分支具备完善的模型训练、预测、推理部署、量化裁剪等功能。 +- develop：基于Paddle静态图开发的分支，推荐使用paddle1.8 或者2.0版本，该分支具备完善的模型训练、预测、推理部署、量化裁剪等功能，领先于release/1.1分支。 - release/1.1：PaddleOCR 发布的第一个稳定版本，基于静态图开发，具备完善的训练、预测、推理部署、量化裁剪等功能。 - dygraph：基于Paddle动态图开发的分支，未来将作为主要开发分支，运行要求使用Paddle2.0rc1版本，目前仍在开发中。 - release/2.0-rc1-0：PaddleOCR发布的第二个稳定版本，基于动态图和paddle2.0版本开发，动态图开发的工程更易于调试，目前支，支持模型训练、预测，暂不支持移动端部署。 @@ -301,7 +301,7 @@ python3 -m pip install paddlepaddle-gpu==2.0.0rc1 -i https://mirror.baidu.com/py #### Q2.3.18: PaddleOCR develop分支和dygraph分支的区别？ **A** 目前PaddleOCR有四个分支，分别是： -- develop：基于Paddle静态图开发的分支，推荐使用paddle1.8 或者2.0版本，该分支具备完善的模型训练、预测、推理部署、量化裁剪等功能。 +- develop：基于Paddle静态图开发的分支，推荐使用paddle1.8 或者2.0版本，该分支具备完善的模型训练、预测、推理部署、量化裁剪等功能，领先于release/1.1分支。 - release/1.1：PaddleOCR 发布的第一个稳定版本，基于静态图开发，具备完善的训练、预测、推理部署、量化裁剪等功能。 - dygraph：基于Paddle动态图开发的分支，目前仍在开发中，未来将作为主要开发分支，运行要求使用Paddle2.0rc1版本，目前仍在开发中。 - release/2.0-rc1-0：PaddleOCR发布的第二个稳定版本，基于动态图和paddle2.0版本开发，动态图开发的工程更易于调试，目前支，支持模型训练、预测，暂不支持移动端部署。 From f0b9aae35333afa96ecd2a332c6d674ce164020d Mon Sep 17 00:00:00 2001 From: LDOUBLEV Date: Mon, 21 Dec 2020 11:49:20 +0800 Subject: [PATCH 34/60] fix comment --- README_ch.md | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/README_ch.md b/README_ch.md index c7ddf0fe..20407d3b 100755 --- a/README_ch.md +++ b/README_ch.md @@ -10,8 +10,8 @@ PaddleOCR同时支持动态图与静态图两种编程范式 **近期更新** - 2020.12.21 [FAQ](./doc/doc_ch/FAQ.md)新增5个高频问题，总数132个，每周一都会更新，欢迎大家持续关注。 - 2020.12.15 更新数据合成工具[Style-Text](./StyleText/README_ch.md)，可以批量合成大量与目标场景类似的图像，在多个场景验证，效果明显提升。 -- 2020.12.14 [FAQ](./doc/doc_ch/FAQ.md)新增5个高频问题，总数127个，每周一都会更新，欢迎大家持续关注。 - 2020.11.25 更新半自动标注工具[PPOCRLabel](./PPOCRLabel/README_ch.md)，辅助开发者高效完成标注任务，输出格式与PP-OCR训练任务完美衔接。 +- 2020.9.22 更新PP-OCR技术文章，https://arxiv.org/abs/2009.09941 - [More](./doc/doc_ch/update.md) @@ -101,8 +101,8 @@ PaddleOCR同时支持动态图与静态图两种编程范式 - [效果展示](#效果展示) - FAQ - [【精选】OCR精选10个问题](./doc/doc_ch/FAQ.md) - - [【理论篇】OCR通用30个问题](./doc/doc_ch/FAQ.md) - - [【实战篇】PaddleOCR实战84个问题](./doc/doc_ch/FAQ.md) + - [【理论篇】OCR通用32个问题](./doc/doc_ch/FAQ.md) + - [【实战篇】PaddleOCR实战90个问题](./doc/doc_ch/FAQ.md) - [技术交流群](#欢迎加入PaddleOCR技术交流群) - [参考文献](./doc/doc_ch/reference.md) - [许可证书](#许可证书) From e1d11ae0e3744f907b0cba4bb3733dc82d7efcb7 Mon Sep 17 00:00:00 2001 From: LDOUBLEV Date: Mon, 21 Dec 2020 15:14:52 +0800 Subject: [PATCH 35/60] 2.3.18 to 3.1.24 --- README_ch.md | 4 ++-- doc/doc_ch/FAQ.md | 57 +++++++++++++++++++++++++---------------------- 2 files changed, 32 insertions(+), 29 deletions(-) diff --git a/README_ch.md b/README_ch.md index 20407d3b..774b074d 100755 --- a/README_ch.md +++ b/README_ch.md @@ -101,8 +101,8 @@ PaddleOCR同时支持动态图与静态图两种编程范式 - [效果展示](#效果展示) - FAQ - [【精选】OCR精选10个问题](./doc/doc_ch/FAQ.md) - - [【理论篇】OCR通用32个问题](./doc/doc_ch/FAQ.md) - - [【实战篇】PaddleOCR实战90个问题](./doc/doc_ch/FAQ.md) + - [【理论篇】OCR通用31个问题](./doc/doc_ch/FAQ.md) + - [【实战篇】PaddleOCR实战91个问题](./doc/doc_ch/FAQ.md) - [技术交流群](#欢迎加入PaddleOCR技术交流群) - [参考文献](./doc/doc_ch/reference.md) - [许可证书](#许可证书) diff --git a/doc/doc_ch/FAQ.md b/doc/doc_ch/FAQ.md index 275d9ac3..288a5492 100755 --- a/doc/doc_ch/FAQ.md +++ b/doc/doc_ch/FAQ.md @@ -11,12 +11,12 @@ * [近期更新（2020.12.21）](#近期更新) * [【精选】OCR精选10个问题](#OCR精选10个问题) -* [【理论篇】OCR通用32个问题](#OCR通用问题) +* [【理论篇】OCR通用31个问题](#OCR通用问题) * [基础知识7题](#基础知识) * [数据集7题](#数据集2) - * [模型训练调优18题](#模型训练调优2) -* [【实战篇】PaddleOCR实战90个问题](#PaddleOCR实战问题) - * [使用咨询23题](#使用咨询) + * [模型训练调优17题](#模型训练调优2) +* [【实战篇】PaddleOCR实战91个问题](#PaddleOCR实战问题) + * [使用咨询24题](#使用咨询) * [数据集17题](#数据集3) * [模型训练调优25题](#模型训练调优3) * [预测部署25题](#预测部署3) @@ -29,17 +29,6 @@ **A**： StyleText模型生成的数据主要用于OCR识别模型的训练。PaddleOCR目前识别模型的输入为32 x N，因此当前版本模型主要适用高度为32的数据。 建议要合成的数据尺寸设置为32 x N。尺寸相差不多的数据也可以生成，尺寸很大或很小的数据效果确实不佳。 -#### Q2.3.18: PaddleOCR develop分支和dygraph分支的区别？ -**A**：目前PaddleOCR有四个分支，分别是： -- develop：基于Paddle静态图开发的分支，推荐使用paddle1.8 或者2.0版本，该分支具备完善的模型训练、预测、推理部署、量化裁剪等功能，领先于release/1.1分支。 -- release/1.1：PaddleOCR 发布的第一个稳定版本，基于静态图开发，具备完善的训练、预测、推理部署、量化裁剪等功能。 -- dygraph：基于Paddle动态图开发的分支，未来将作为主要开发分支，运行要求使用Paddle2.0rc1版本，目前仍在开发中。 -- release/2.0-rc1-0：PaddleOCR发布的第二个稳定版本，基于动态图和paddle2.0版本开发，动态图开发的工程更易于调试，目前支，支持模型训练、预测，暂不支持移动端部署。 - -如果您已经上手过PaddleOCR，并且希望在各种环境上部署PaddleOCR，目前建议使用静态图分支，develop或者release/1.1分支。如果您是初学者，想快速训练，调试PaddleOCR中的算法，建议尝鲜PaddleOCR dygraph分支。 - -**注意**：develop和dygraph分支要求的Paddle版本、本地环境有差别，请注意不同分支环境安装部分的差异。 - #### Q3.1.22： ModuleNotFoundError: No module named 'paddle.nn'， **A**： paddle.nn是Paddle2.0版本特有的功能，请安装大于等于Paddle 2.0.0rc1的版本，安装方式为 ``` @@ -54,6 +43,18 @@ python3 -m pip install paddlepaddle-gpu==2.0.0rc1 -i https://mirror.baidu.com/py 2.0版本推荐使用新的docker镜像 `paddlepaddle/paddle:latest-dev-cuda10.1-cudnn7-gcc82`。 或者访问[dockerhub](https://hub.docker.com/r/paddlepaddle/paddle/tags/)获得与您机器适配的镜像。 +#### Q3.1.24: PaddleOCR develop分支和dygraph分支的区别？ +**A**：目前PaddleOCR有四个分支，分别是： +- develop：基于Paddle静态图开发的分支，推荐使用paddle1.8 或者2.0版本，该分支具备完善的模型训练、预测、推理部署、量化裁剪等功能，领先于release/1.1分支。 +- release/1.1：PaddleOCR 发布的第一个稳定版本，基于静态图开发，具备完善的训练、预测、推理部署、量化裁剪等功能。 +- dygraph：基于Paddle动态图开发的分支，未来将作为主要开发分支，运行要求使用Paddle2.0rc1版本，目前仍在开发中。 +- release/2.0-rc1-0：PaddleOCR发布的第二个稳定版本，基于动态图和paddle2.0版本开发，动态图开发的工程更易于调试，目前支，支持模型训练、预测，暂不支持移动端部署。 + +如果您已经上手过PaddleOCR，并且希望在各种环境上部署PaddleOCR，目前建议使用静态图分支，develop或者release/1.1分支。如果您是初学者，想快速训练，调试PaddleOCR中的算法，建议尝鲜PaddleOCR dygraph分支。 + +**注意**：develop和dygraph分支要求的Paddle版本、本地环境有差别，请注意不同分支环境安装部分的差异。 + + #### Q3.4.25 : PaddleOCR模型Python端预测和C++预测结果不一致？ **A**：正常来说，python端预测和C++预测文本是一致的，如果预测结果差异较大，建议首先排查diff出现在检测模型还是识别模型，或者尝试换其他模型是否有类似的问题。其次，检查python端和C++端数据处理部分是否存在差异，建议保存环境，更新PaddleOCR代码再试下。如果更新代码或者更新代码都没能解决，建议在PaddleOCR群里或者issue中抛出您的问题。 @@ -298,18 +299,6 @@ python3 -m pip install paddlepaddle-gpu==2.0.0rc1 -i https://mirror.baidu.com/py **A**：StyleText模型生成的数据主要用于OCR识别模型的训练。PaddleOCR目前识别模型的输入为32 x N，因此当前版本模型主要适用高度为32的数据。 建议要合成的数据尺寸设置为32 x N。尺寸相差不多的数据也可以生成，尺寸很大或很小的数据效果确实不佳。 -#### Q2.3.18: PaddleOCR develop分支和dygraph分支的区别？ -**A** 目前PaddleOCR有四个分支，分别是： - -- develop：基于Paddle静态图开发的分支，推荐使用paddle1.8 或者2.0版本，该分支具备完善的模型训练、预测、推理部署、量化裁剪等功能，领先于release/1.1分支。 -- release/1.1：PaddleOCR 发布的第一个稳定版本，基于静态图开发，具备完善的训练、预测、推理部署、量化裁剪等功能。 -- dygraph：基于Paddle动态图开发的分支，目前仍在开发中，未来将作为主要开发分支，运行要求使用Paddle2.0rc1版本，目前仍在开发中。 -- release/2.0-rc1-0：PaddleOCR发布的第二个稳定版本，基于动态图和paddle2.0版本开发，动态图开发的工程更易于调试，目前支，支持模型训练、预测，暂不支持移动端部署。 - -如果您已经上手过PaddleOCR，并且希望在各种环境上部署PaddleOCR，目前建议使用静态图分支，develop或者release/1.1分支。如果您是初学者，想快速训练，调试PaddleOCR中的算法，建议尝鲜PaddleOCR dygraph分支。 - -**注意**：develop和dygraph分支要求的Paddle版本、本地环境有差别，请注意不同分支环境安装部分的差异。 - ## 【实战篇】PaddleOCR实战问题 @@ -430,6 +419,20 @@ python3 -m pip install paddlepaddle-gpu==2.0.0rc1 -i https://mirror.baidu.com/py 或者访问[dockerhub](https://hub.docker.com/r/paddlepaddle/paddle/tags/)获得与您机器适配的镜像。 +#### Q3.1.24: PaddleOCR develop分支和dygraph分支的区别？ +**A** 目前PaddleOCR有四个分支，分别是： + +- develop：基于Paddle静态图开发的分支，推荐使用paddle1.8 或者2.0版本，该分支具备完善的模型训练、预测、推理部署、量化裁剪等功能，领先于release/1.1分支。 +- release/1.1：PaddleOCR 发布的第一个稳定版本，基于静态图开发，具备完善的训练、预测、推理部署、量化裁剪等功能。 +- dygraph：基于Paddle动态图开发的分支，目前仍在开发中，未来将作为主要开发分支，运行要求使用Paddle2.0rc1版本，目前仍在开发中。 +- release/2.0-rc1-0：PaddleOCR发布的第二个稳定版本，基于动态图和paddle2.0版本开发，动态图开发的工程更易于调试，目前支，支持模型训练、预测，暂不支持移动端部署。 + +如果您已经上手过PaddleOCR，并且希望在各种环境上部署PaddleOCR，目前建议使用静态图分支，develop或者release/1.1分支。如果您是初学者，想快速训练，调试PaddleOCR中的算法，建议尝鲜PaddleOCR dygraph分支。 + +**注意**：develop和dygraph分支要求的Paddle版本、本地环境有差别，请注意不同分支环境安装部分的差异。 + + + ### 数据集 From 390f240b822aac4f1e2b25a5eb5659ff11c9259a Mon Sep 17 00:00:00 2001 From: LDOUBLEV Date: Mon, 21 Dec 2020 16:30:04 +0800 Subject: [PATCH 36/60] python to python3 --- StyleText/README.md | 4 ++-- StyleText/README_ch.md | 2 +- 2 files changed, 3 insertions(+), 3 deletions(-) diff --git a/StyleText/README.md b/StyleText/README.md index f978f438..5262b505 100644 --- a/StyleText/README.md +++ b/StyleText/README.md @@ -119,7 +119,7 @@ In actual application scenarios, it is often necessary to synthesize pictures in * `corpus_file`: Filepath of the corpus. Corpus file should be a text file which will be split by line-endings（'\n'）. Corpus generator samples one line each time. -Example of corpus file: +Example of corpus file: ``` PaddleOCR 飞桨文字识别 @@ -136,7 +136,7 @@ We provide a general dataset containing Chinese, English and Korean (50,000 imag 2. You can run the following command to start synthesis task: ``` bash - python -m tools.synth_dataset.py -c configs/dataset_config.yml + python3 -m tools.synth_dataset.py -c configs/dataset_config.yml ``` diff --git a/StyleText/README_ch.md b/StyleText/README_ch.md index a8ab933b..cce136c0 100644 --- a/StyleText/README_ch.md +++ b/StyleText/README_ch.md @@ -124,7 +124,7 @@ python3 -m tools.synth_image -c configs/config.yml --style_image examples/style_ 2. 运行`tools/synth_dataset`合成数据： ``` bash - python -m tools.synth_dataset -c configs/dataset_config.yml + python3 -m tools.synth_dataset -c configs/dataset_config.yml ``` From 59af7359befb7ba50109e3f694f740a064685a95 Mon Sep 17 00:00:00 2001 From: WenmuZhou Date: Mon, 21 Dec 2020 17:10:00 +0800 Subject: [PATCH 37/60] inference adaptation 2.0 --- tools/infer/predict_cls.py | 14 +++++--------- tools/infer/predict_det.py | 13 +++++-------- tools/infer/predict_rec.py | 13 ++++--------- tools/infer/utility.py | 25 ++++++++++--------------- 4 files changed, 24 insertions(+), 41 deletions(-) diff --git a/tools/infer/predict_cls.py b/tools/infer/predict_cls.py index 420213ee..3bda7d7d 100755 --- a/tools/infer/predict_cls.py +++ b/tools/infer/predict_cls.py @@ -39,7 +39,6 @@ class TextClassifier(object): self.cls_image_shape = [int(v) for v in args.cls_image_shape.split(",")] self.cls_batch_num = args.cls_batch_num self.cls_thresh = args.cls_thresh - self.use_zero_copy_run = args.use_zero_copy_run postprocess_params = { 'name': 'ClsPostProcess', "label_list": args.label_list, @@ -99,12 +98,8 @@ class TextClassifier(object): norm_img_batch = norm_img_batch.copy() starttime = time.time() - if self.use_zero_copy_run: - self.input_tensor.copy_from_cpu(norm_img_batch) - self.predictor.zero_copy_run() - else: - norm_img_batch = fluid.core.PaddleTensor(norm_img_batch) - self.predictor.run([norm_img_batch]) + self.input_tensor.copy_from_cpu(norm_img_batch) + self.predictor.run() prob_out = self.output_tensors[0].copy_to_cpu() cls_result = self.postprocess_op(prob_out) elapse += time.time() - starttime @@ -143,10 +138,11 @@ def main(args): "Please set --rec_image_shape='3,32,100' and --rec_char_type='en' ") exit() for ino in range(len(img_list)): - logger.info("Predicts of {}:{}".format(valid_image_file_list[ino], cls_res[ - ino])) + logger.info("Predicts of {}:{}".format(valid_image_file_list[ino], + cls_res[ino])) logger.info("Total predict time for {} images, cost: {:.3f}".format( len(img_list), predict_time)) + if __name__ == "__main__": main(utility.parse_args()) diff --git a/tools/infer/predict_det.py b/tools/infer/predict_det.py index fe772991..ad9cbf57 100755 --- a/tools/infer/predict_det.py +++ b/tools/infer/predict_det.py @@ -37,7 +37,6 @@ class TextDetector(object): def __init__(self, args): self.args = args self.det_algorithm = args.det_algorithm - self.use_zero_copy_run = args.use_zero_copy_run pre_process_list = [{ 'DetResizeForTest': { 'limit_side_len': args.det_limit_side_len, @@ -72,7 +71,9 @@ class TextDetector(object): postprocess_params["nms_thresh"] = args.det_east_nms_thresh elif self.det_algorithm == "SAST": pre_process_list[0] = { - 'DetResizeForTest': {'resize_long': args.det_limit_side_len} + 'DetResizeForTest': { + 'resize_long': args.det_limit_side_len + } } postprocess_params['name'] = 'SASTPostProcess' postprocess_params["score_thresh"] = args.det_sast_score_thresh @@ -161,12 +162,8 @@ class TextDetector(object): img = img.copy() starttime = time.time() - if self.use_zero_copy_run: - self.input_tensor.copy_from_cpu(img) - self.predictor.zero_copy_run() - else: - im = paddle.fluid.core.PaddleTensor(img) - self.predictor.run([im]) + self.input_tensor.copy_from_cpu(img) + self.predictor.run() outputs = [] for output_tensor in self.output_tensors: output = output_tensor.copy_to_cpu() diff --git a/tools/infer/predict_rec.py b/tools/infer/predict_rec.py index c615fa0d..54082a50 100755 --- a/tools/infer/predict_rec.py +++ b/tools/infer/predict_rec.py @@ -39,7 +39,6 @@ class TextRecognizer(object): self.character_type = args.rec_char_type self.rec_batch_num = args.rec_batch_num self.rec_algorithm = args.rec_algorithm - self.use_zero_copy_run = args.use_zero_copy_run postprocess_params = { 'name': 'CTCLabelDecode', "character_type": args.rec_char_type, @@ -101,12 +100,8 @@ class TextRecognizer(object): norm_img_batch = np.concatenate(norm_img_batch) norm_img_batch = norm_img_batch.copy() starttime = time.time() - if self.use_zero_copy_run: - self.input_tensor.copy_from_cpu(norm_img_batch) - self.predictor.zero_copy_run() - else: - norm_img_batch = fluid.core.PaddleTensor(norm_img_batch) - self.predictor.run([norm_img_batch]) + self.input_tensor.copy_from_cpu(norm_img_batch) + self.predictor.run() outputs = [] for output_tensor in self.output_tensors: output = output_tensor.copy_to_cpu() @@ -145,8 +140,8 @@ def main(args): "Please set --rec_image_shape='3,32,100' and --rec_char_type='en' ") exit() for ino in range(len(img_list)): - logger.info("Predicts of {}:{}".format(valid_image_file_list[ino], rec_res[ - ino])) + logger.info("Predicts of {}:{}".format(valid_image_file_list[ino], + rec_res[ino])) logger.info("Total predict time for {} images, cost: {:.3f}".format( len(img_list), predict_time)) diff --git a/tools/infer/utility.py b/tools/infer/utility.py index c3d294e6..39c045de 100755 --- a/tools/infer/utility.py +++ b/tools/infer/utility.py @@ -20,8 +20,7 @@ import numpy as np import json from PIL import Image, ImageDraw, ImageFont import math -from paddle.fluid.core import AnalysisConfig -from paddle.fluid.core import create_paddle_predictor +from paddle import inference def parse_args(): @@ -83,8 +82,6 @@ def parse_args(): parser.add_argument("--cls_thresh", type=float, default=0.9) parser.add_argument("--enable_mkldnn", type=str2bool, default=False) - parser.add_argument("--use_zero_copy_run", type=str2bool, default=False) - parser.add_argument("--use_pdserving", type=str2bool, default=False) return parser.parse_args() @@ -110,14 +107,14 @@ def create_predictor(args, mode, logger): logger.info("not find params file path {}".format(params_file_path)) sys.exit(0) - config = AnalysisConfig(model_file_path, params_file_path) + config = inference.Config(model_file_path, params_file_path) if args.use_gpu: config.enable_use_gpu(args.gpu_mem, 0) if args.use_tensorrt: config.enable_tensorrt_engine( - precision_mode=AnalysisConfig.Precision.Half - if args.use_fp16 else AnalysisConfig.Precision.Float32, + precision_mode=inference.PrecisionType.Half + if args.use_fp16 else inference.PrecisionType.Float32, max_batch_size=args.max_batch_size) else: config.disable_gpu() @@ -130,20 +127,18 @@ def create_predictor(args, mode, logger): # config.enable_memory_optim() config.disable_glog_info() - if args.use_zero_copy_run: - config.delete_pass("conv_transpose_eltwiseadd_bn_fuse_pass") - config.switch_use_feed_fetch_ops(False) - else: - config.switch_use_feed_fetch_ops(True) + config.delete_pass("conv_transpose_eltwiseadd_bn_fuse_pass") + config.switch_use_feed_fetch_ops(False) - predictor = create_paddle_predictor(config) + # create predictor + predictor = inference.create_predictor(config) input_names = predictor.get_input_names() for name in input_names: - input_tensor = predictor.get_input_tensor(name) + input_tensor = predictor.get_input_handle(name) output_names = predictor.get_output_names() output_tensors = [] for output_name in output_names: - output_tensor = predictor.get_output_tensor(output_name) + output_tensor = predictor.get_output_handle(output_name) output_tensors.append(output_tensor) return predictor, input_tensor, output_tensors From 3a0090dc004a1411c4d5be30892c62c1a255b204 Mon Sep 17 00:00:00 2001 From: WenmuZhou Date: Mon, 21 Dec 2020 17:13:32 +0800 Subject: [PATCH 38/60] delete fluid --- ppocr/losses/det_sast_loss.py | 44 +++++++++++++++++------------------ tools/infer/predict_cls.py | 1 - tools/infer/predict_det.py | 1 - tools/infer/predict_rec.py | 1 - tools/program.py | 2 +- 5 files changed, 23 insertions(+), 26 deletions(-) diff --git a/ppocr/losses/det_sast_loss.py b/ppocr/losses/det_sast_loss.py index a07af6a4..2e0c756b 100644 --- a/ppocr/losses/det_sast_loss.py +++ b/ppocr/losses/det_sast_loss.py @@ -19,7 +19,6 @@ from __future__ import print_function import paddle from paddle import nn from .det_basic_loss import DiceLoss -import paddle.fluid as fluid import numpy as np @@ -27,9 +26,7 @@ class SASTLoss(nn.Layer): """ """ - def __init__(self, - eps=1e-6, - **kwargs): + def __init__(self, eps=1e-6, **kwargs): super(SASTLoss, self).__init__() self.dice_loss = DiceLoss(eps=eps) @@ -39,7 +36,7 @@ class SASTLoss(nn.Layer): tcl_mask: N x 128 x 1 tcl_label: N x X list or LoDTensor """ - + f_score = predicts['f_score'] f_border = predicts['f_border'] f_tvo = predicts['f_tvo'] @@ -53,15 +50,17 @@ class SASTLoss(nn.Layer): score_loss = 1.0 - 2 * intersection / (union + 1e-5) #border loss - l_border_split, l_border_norm = paddle.split(l_border, num_or_sections=[4, 1], axis=1) + l_border_split, l_border_norm = paddle.split( + l_border, num_or_sections=[4, 1], axis=1) f_border_split = f_border border_ex_shape = l_border_norm.shape * np.array([1, 4, 1, 1]) - l_border_norm_split = paddle.expand(x=l_border_norm, shape=border_ex_shape) - l_border_score = paddle.expand(x=l_score, shape=border_ex_shape) - l_border_mask = paddle.expand(x=l_mask, shape=border_ex_shape) + l_border_norm_split = paddle.expand( + x=l_border_norm, shape=border_ex_shape) + l_border_score = paddle.expand(x=l_score, shape=border_ex_shape) + l_border_mask = paddle.expand(x=l_mask, shape=border_ex_shape) border_diff = l_border_split - f_border_split - abs_border_diff = paddle.abs(border_diff) + abs_border_diff = paddle.abs(border_diff) border_sign = abs_border_diff < 1.0 border_sign = paddle.cast(border_sign, dtype='float32') border_sign.stop_gradient = True @@ -72,15 +71,16 @@ class SASTLoss(nn.Layer): (paddle.sum(l_border_score * l_border_mask) + 1e-5) #tvo_loss - l_tvo_split, l_tvo_norm = paddle.split(l_tvo, num_or_sections=[8, 1], axis=1) + l_tvo_split, l_tvo_norm = paddle.split( + l_tvo, num_or_sections=[8, 1], axis=1) f_tvo_split = f_tvo tvo_ex_shape = l_tvo_norm.shape * np.array([1, 8, 1, 1]) l_tvo_norm_split = paddle.expand(x=l_tvo_norm, shape=tvo_ex_shape) - l_tvo_score = paddle.expand(x=l_score, shape=tvo_ex_shape) - l_tvo_mask = paddle.expand(x=l_mask, shape=tvo_ex_shape) + l_tvo_score = paddle.expand(x=l_score, shape=tvo_ex_shape) + l_tvo_mask = paddle.expand(x=l_mask, shape=tvo_ex_shape) # tvo_geo_diff = l_tvo_split - f_tvo_split - abs_tvo_geo_diff = paddle.abs(tvo_geo_diff) + abs_tvo_geo_diff = paddle.abs(tvo_geo_diff) tvo_sign = abs_tvo_geo_diff < 1.0 tvo_sign = paddle.cast(tvo_sign, dtype='float32') tvo_sign.stop_gradient = True @@ -91,15 +91,16 @@ class SASTLoss(nn.Layer): (paddle.sum(l_tvo_score * l_tvo_mask) + 1e-5) #tco_loss - l_tco_split, l_tco_norm = paddle.split(l_tco, num_or_sections=[2, 1], axis=1) + l_tco_split, l_tco_norm = paddle.split( + l_tco, num_or_sections=[2, 1], axis=1) f_tco_split = f_tco tco_ex_shape = l_tco_norm.shape * np.array([1, 2, 1, 1]) l_tco_norm_split = paddle.expand(x=l_tco_norm, shape=tco_ex_shape) - l_tco_score = paddle.expand(x=l_score, shape=tco_ex_shape) - l_tco_mask = paddle.expand(x=l_mask, shape=tco_ex_shape) - + l_tco_score = paddle.expand(x=l_score, shape=tco_ex_shape) + l_tco_mask = paddle.expand(x=l_mask, shape=tco_ex_shape) + tco_geo_diff = l_tco_split - f_tco_split - abs_tco_geo_diff = paddle.abs(tco_geo_diff) + abs_tco_geo_diff = paddle.abs(tco_geo_diff) tco_sign = abs_tco_geo_diff < 1.0 tco_sign = paddle.cast(tco_sign, dtype='float32') tco_sign.stop_gradient = True @@ -109,13 +110,12 @@ class SASTLoss(nn.Layer): tco_loss = paddle.sum(tco_out_loss * l_tco_score * l_tco_mask) / \ (paddle.sum(l_tco_score * l_tco_mask) + 1e-5) - # total loss tvo_lw, tco_lw = 1.5, 1.5 score_lw, border_lw = 1.0, 1.0 total_loss = score_loss * score_lw + border_loss * border_lw + \ tvo_loss * tvo_lw + tco_loss * tco_lw - + losses = {'loss':total_loss, "score_loss":score_loss,\ "border_loss":border_loss, 'tvo_loss':tvo_loss, 'tco_loss':tco_loss} - return losses \ No newline at end of file + return losses diff --git a/tools/infer/predict_cls.py b/tools/infer/predict_cls.py index 3bda7d7d..6604d3cc 100755 --- a/tools/infer/predict_cls.py +++ b/tools/infer/predict_cls.py @@ -24,7 +24,6 @@ import numpy as np import math import time import traceback -import paddle.fluid as fluid import tools.infer.utility as utility from ppocr.postprocess import build_post_process diff --git a/tools/infer/predict_det.py b/tools/infer/predict_det.py index ad9cbf57..e677eb24 100755 --- a/tools/infer/predict_det.py +++ b/tools/infer/predict_det.py @@ -22,7 +22,6 @@ import cv2 import numpy as np import time import sys -import paddle import tools.infer.utility as utility from ppocr.utils.logging import get_logger diff --git a/tools/infer/predict_rec.py b/tools/infer/predict_rec.py index 54082a50..89d04cff 100755 --- a/tools/infer/predict_rec.py +++ b/tools/infer/predict_rec.py @@ -23,7 +23,6 @@ import numpy as np import math import time import traceback -import paddle.fluid as fluid import tools.infer.utility as utility from ppocr.postprocess import build_post_process diff --git a/tools/program.py b/tools/program.py index 4331f9d4..c712fe14 100755 --- a/tools/program.py +++ b/tools/program.py @@ -131,7 +131,7 @@ def check_gpu(use_gpu): "model on CPU" try: - if use_gpu and not paddle.fluid.is_compiled_with_cuda(): + if use_gpu and not paddle.is_compiled_with_cuda(): print(err) sys.exit(1) except Exception as e: From 551377a013bfbce830d16cf445248a1114c63e32 Mon Sep 17 00:00:00 2001 From: LDOUBLEV Date: Mon, 21 Dec 2020 13:41:33 +0000 Subject: [PATCH 39/60] update cpp_infer to 2.0 --- deploy/cpp_infer/include/config.h | 8 ++---- deploy/cpp_infer/include/ocr_cls.h | 9 +++---- deploy/cpp_infer/include/ocr_det.h | 10 ++++---- deploy/cpp_infer/include/ocr_rec.h | 9 +++---- deploy/cpp_infer/src/config.cpp | 8 +++--- deploy/cpp_infer/src/main.cpp | 18 ++++++-------- deploy/cpp_infer/src/ocr_cls.cpp | 28 +++++++-------------- deploy/cpp_infer/src/ocr_det.cpp | 39 ++++++++++++------------------ deploy/cpp_infer/src/ocr_rec.cpp | 33 +++++++++---------------- deploy/cpp_infer/tools/config.txt | 7 +++--- 10 files changed, 66 insertions(+), 103 deletions(-) diff --git a/deploy/cpp_infer/include/config.h b/deploy/cpp_infer/include/config.h index 27539ea7..3faeede1 100644 --- a/deploy/cpp_infer/include/config.h +++ b/deploy/cpp_infer/include/config.h @@ -25,9 +25,9 @@ namespace PaddleOCR { -class Config { +class OCRConfig { public: - explicit Config(const std::string &config_file) { + explicit OCRConfig(const std::string &config_file) { config_map_ = LoadConfig(config_file); this->use_gpu = bool(stoi(config_map_["use_gpu"])); @@ -41,8 +41,6 @@ public: this->use_mkldnn = bool(stoi(config_map_["use_mkldnn"])); - this->use_zero_copy_run = bool(stoi(config_map_["use_zero_copy_run"])); - this->max_side_len = stoi(config_map_["max_side_len"]); this->det_db_thresh = stod(config_map_["det_db_thresh"]); @@ -76,8 +74,6 @@ public: bool use_mkldnn = false; - bool use_zero_copy_run = false; - int max_side_len = 960; double det_db_thresh = 0.3; diff --git a/deploy/cpp_infer/include/ocr_cls.h b/deploy/cpp_infer/include/ocr_cls.h index 38a37cff..87772cc1 100644 --- a/deploy/cpp_infer/include/ocr_cls.h +++ b/deploy/cpp_infer/include/ocr_cls.h @@ -30,6 +30,8 @@ #include #include +using namespace paddle_infer; + namespace PaddleOCR { class Classifier { @@ -37,14 +39,12 @@ public: explicit Classifier(const std::string &model_dir, const bool &use_gpu, const int &gpu_id, const int &gpu_mem, const int &cpu_math_library_num_threads, - const bool &use_mkldnn, const bool &use_zero_copy_run, - const double &cls_thresh) { + const bool &use_mkldnn, const double &cls_thresh) { this->use_gpu_ = use_gpu; this->gpu_id_ = gpu_id; this->gpu_mem_ = gpu_mem; this->cpu_math_library_num_threads_ = cpu_math_library_num_threads; this->use_mkldnn_ = use_mkldnn; - this->use_zero_copy_run_ = use_zero_copy_run; this->cls_thresh = cls_thresh; @@ -57,14 +57,13 @@ public: cv::Mat Run(cv::Mat &img); private: - std::shared_ptr predictor_; + std::shared_ptr predictor_; bool use_gpu_ = false; int gpu_id_ = 0; int gpu_mem_ = 4000; int cpu_math_library_num_threads_ = 4; bool use_mkldnn_ = false; - bool use_zero_copy_run_ = false; double cls_thresh = 0.5; std::vector mean_ = {0.5f, 0.5f, 0.5f}; diff --git a/deploy/cpp_infer/include/ocr_det.h b/deploy/cpp_infer/include/ocr_det.h index 0308d07f..d50fd70a 100644 --- a/deploy/cpp_infer/include/ocr_det.h +++ b/deploy/cpp_infer/include/ocr_det.h @@ -32,6 +32,8 @@ #include #include +using namespace paddle_infer; + namespace PaddleOCR { class DBDetector { @@ -39,8 +41,8 @@ public: explicit DBDetector(const std::string &model_dir, const bool &use_gpu, const int &gpu_id, const int &gpu_mem, const int &cpu_math_library_num_threads, - const bool &use_mkldnn, const bool &use_zero_copy_run, - const int &max_side_len, const double &det_db_thresh, + const bool &use_mkldnn, const int &max_side_len, + const double &det_db_thresh, const double &det_db_box_thresh, const double &det_db_unclip_ratio, const bool &visualize) { @@ -49,7 +51,6 @@ public: this->gpu_mem_ = gpu_mem; this->cpu_math_library_num_threads_ = cpu_math_library_num_threads; this->use_mkldnn_ = use_mkldnn; - this->use_zero_copy_run_ = use_zero_copy_run; this->max_side_len_ = max_side_len; @@ -69,14 +70,13 @@ public: void Run(cv::Mat &img, std::vector>> &boxes); private: - std::shared_ptr predictor_; + std::shared_ptr predictor_; bool use_gpu_ = false; int gpu_id_ = 0; int gpu_mem_ = 4000; int cpu_math_library_num_threads_ = 4; bool use_mkldnn_ = false; - bool use_zero_copy_run_ = false; int max_side_len_ = 960; diff --git a/deploy/cpp_infer/include/ocr_rec.h b/deploy/cpp_infer/include/ocr_rec.h index 89bcd82c..14b77b08 100644 --- a/deploy/cpp_infer/include/ocr_rec.h +++ b/deploy/cpp_infer/include/ocr_rec.h @@ -32,6 +32,8 @@ #include #include +using namespace paddle_infer; + namespace PaddleOCR { class CRNNRecognizer { @@ -39,14 +41,12 @@ public: explicit CRNNRecognizer(const std::string &model_dir, const bool &use_gpu, const int &gpu_id, const int &gpu_mem, const int &cpu_math_library_num_threads, - const bool &use_mkldnn, const bool &use_zero_copy_run, - const string &label_path) { + const bool &use_mkldnn, const string &label_path) { this->use_gpu_ = use_gpu; this->gpu_id_ = gpu_id; this->gpu_mem_ = gpu_mem; this->cpu_math_library_num_threads_ = cpu_math_library_num_threads; this->use_mkldnn_ = use_mkldnn; - this->use_zero_copy_run_ = use_zero_copy_run; this->label_list_ = Utility::ReadDict(label_path); this->label_list_.insert(this->label_list_.begin(), @@ -63,14 +63,13 @@ public: Classifier *cls); private: - std::shared_ptr predictor_; + std::shared_ptr predictor_; bool use_gpu_ = false; int gpu_id_ = 0; int gpu_mem_ = 4000; int cpu_math_library_num_threads_ = 4; bool use_mkldnn_ = false; - bool use_zero_copy_run_ = false; std::vector label_list_; diff --git a/deploy/cpp_infer/src/config.cpp b/deploy/cpp_infer/src/config.cpp index 52dfa209..303c3c12 100644 --- a/deploy/cpp_infer/src/config.cpp +++ b/deploy/cpp_infer/src/config.cpp @@ -16,8 +16,8 @@ namespace PaddleOCR { -std::vector Config::split(const std::string &str, - const std::string &delim) { +std::vector OCRConfig::split(const std::string &str, + const std::string &delim) { std::vector res; if ("" == str) return res; @@ -38,7 +38,7 @@ std::vector Config::split(const std::string &str, } std::map -Config::LoadConfig(const std::string &config_path) { +OCRConfig::LoadConfig(const std::string &config_path) { auto config = Utility::ReadDict(config_path); std::map dict; @@ -53,7 +53,7 @@ Config::LoadConfig(const std::string &config_path) { return dict; } -void Config::PrintConfigInfo() { +void OCRConfig::PrintConfigInfo() { std::cout << "=======Paddle OCR inference config======" << std::endl; for (auto iter = config_map_.begin(); iter != config_map_.end(); iter++) { std::cout << iter->first << " : " << iter->second << std::endl; diff --git a/deploy/cpp_infer/src/main.cpp b/deploy/cpp_infer/src/main.cpp index 63da62c7..187b70dc 100644 --- a/deploy/cpp_infer/src/main.cpp +++ b/deploy/cpp_infer/src/main.cpp @@ -42,7 +42,7 @@ int main(int argc, char **argv) { exit(1); } - Config config(argv[1]); + OCRConfig config(argv[1]); config.PrintConfigInfo(); @@ -50,24 +50,22 @@ int main(int argc, char **argv) { cv::Mat srcimg = cv::imread(img_path, cv::IMREAD_COLOR); - DBDetector det( - config.det_model_dir, config.use_gpu, config.gpu_id, config.gpu_mem, - config.cpu_math_library_num_threads, config.use_mkldnn, - config.use_zero_copy_run, config.max_side_len, config.det_db_thresh, - config.det_db_box_thresh, config.det_db_unclip_ratio, config.visualize); + DBDetector det(config.det_model_dir, config.use_gpu, config.gpu_id, + config.gpu_mem, config.cpu_math_library_num_threads, + config.use_mkldnn, config.max_side_len, config.det_db_thresh, + config.det_db_box_thresh, config.det_db_unclip_ratio, + config.visualize); Classifier *cls = nullptr; if (config.use_angle_cls == true) { cls = new Classifier(config.cls_model_dir, config.use_gpu, config.gpu_id, config.gpu_mem, config.cpu_math_library_num_threads, - config.use_mkldnn, config.use_zero_copy_run, - config.cls_thresh); + config.use_mkldnn, config.cls_thresh); } CRNNRecognizer rec(config.rec_model_dir, config.use_gpu, config.gpu_id, config.gpu_mem, config.cpu_math_library_num_threads, - config.use_mkldnn, config.use_zero_copy_run, - config.char_list_file); + config.use_mkldnn, config.char_list_file); #ifdef USE_MKL #pragma omp parallel diff --git a/deploy/cpp_infer/src/ocr_cls.cpp b/deploy/cpp_infer/src/ocr_cls.cpp index fed2023f..9757b482 100644 --- a/deploy/cpp_infer/src/ocr_cls.cpp +++ b/deploy/cpp_infer/src/ocr_cls.cpp @@ -35,26 +35,16 @@ cv::Mat Classifier::Run(cv::Mat &img) { this->permute_op_.Run(&resize_img, input.data()); // Inference. - if (this->use_zero_copy_run_) { - auto input_names = this->predictor_->GetInputNames(); - auto input_t = this->predictor_->GetInputTensor(input_names[0]); - input_t->Reshape({1, 3, resize_img.rows, resize_img.cols}); - input_t->copy_from_cpu(input.data()); - this->predictor_->ZeroCopyRun(); - } else { - paddle::PaddleTensor input_t; - input_t.shape = {1, 3, resize_img.rows, resize_img.cols}; - input_t.data = - paddle::PaddleBuf(input.data(), input.size() * sizeof(float)); - input_t.dtype = PaddleDType::FLOAT32; - std::vector outputs; - this->predictor_->Run({input_t}, &outputs, 1); - } + auto input_names = this->predictor_->GetInputNames(); + auto input_t = this->predictor_->GetInputHandle(input_names[0]); + input_t->Reshape({1, 3, resize_img.rows, resize_img.cols}); + input_t->CopyFromCpu(input.data()); + this->predictor_->Run(); std::vector softmax_out; std::vector label_out; auto output_names = this->predictor_->GetOutputNames(); - auto softmax_out_t = this->predictor_->GetOutputTensor(output_names[0]); + auto softmax_out_t = this->predictor_->GetOutputHandle(output_names[0]); auto softmax_shape_out = softmax_out_t->shape(); int softmax_out_num = @@ -63,7 +53,7 @@ cv::Mat Classifier::Run(cv::Mat &img) { softmax_out.resize(softmax_out_num); - softmax_out_t->copy_to_cpu(softmax_out.data()); + softmax_out_t->CopyToCpu(softmax_out.data()); float score = 0; int label = 0; @@ -95,7 +85,7 @@ void Classifier::LoadModel(const std::string &model_dir) { } // false for zero copy tensor - config.SwitchUseFeedFetchOps(!this->use_zero_copy_run_); + config.SwitchUseFeedFetchOps(false); // true for multiple input config.SwitchSpecifyInputNames(true); @@ -104,6 +94,6 @@ void Classifier::LoadModel(const std::string &model_dir) { config.EnableMemoryOptim(); config.DisableGlogInfo(); - this->predictor_ = CreatePaddlePredictor(config); + this->predictor_ = CreatePredictor(config); } } // namespace PaddleOCR diff --git a/deploy/cpp_infer/src/ocr_det.cpp b/deploy/cpp_infer/src/ocr_det.cpp index e253f9cc..c6c93991 100644 --- a/deploy/cpp_infer/src/ocr_det.cpp +++ b/deploy/cpp_infer/src/ocr_det.cpp @@ -17,12 +17,17 @@ namespace PaddleOCR { void DBDetector::LoadModel(const std::string &model_dir) { - AnalysisConfig config; + // AnalysisConfig config; + paddle_infer::Config config; config.SetModel(model_dir + "/inference.pdmodel", model_dir + "/inference.pdiparams"); if (this->use_gpu_) { config.EnableUseGpu(this->gpu_mem_, this->gpu_id_); + // config.EnableTensorRtEngine( + // 1 << 20, 1, 3, + // AnalysisConfig::Precision::kFloat32, + // false, false); } else { config.DisableGpu(); if (this->use_mkldnn_) { @@ -32,10 +37,8 @@ void DBDetector::LoadModel(const std::string &model_dir) { } config.SetCpuMathLibraryNumThreads(this->cpu_math_library_num_threads_); } - - // false for zero copy tensor - // true for commom tensor - config.SwitchUseFeedFetchOps(!this->use_zero_copy_run_); + // use zero_copy_run as default + config.SwitchUseFeedFetchOps(false); // true for multiple input config.SwitchSpecifyInputNames(true); @@ -44,7 +47,7 @@ void DBDetector::LoadModel(const std::string &model_dir) { config.EnableMemoryOptim(); config.DisableGlogInfo(); - this->predictor_ = CreatePaddlePredictor(config); + this->predictor_ = CreatePredictor(config); } void DBDetector::Run(cv::Mat &img, @@ -64,31 +67,21 @@ void DBDetector::Run(cv::Mat &img, this->permute_op_.Run(&resize_img, input.data()); // Inference. - if (this->use_zero_copy_run_) { - auto input_names = this->predictor_->GetInputNames(); - auto input_t = this->predictor_->GetInputTensor(input_names[0]); - input_t->Reshape({1, 3, resize_img.rows, resize_img.cols}); - input_t->copy_from_cpu(input.data()); - this->predictor_->ZeroCopyRun(); - } else { - paddle::PaddleTensor input_t; - input_t.shape = {1, 3, resize_img.rows, resize_img.cols}; - input_t.data = - paddle::PaddleBuf(input.data(), input.size() * sizeof(float)); - input_t.dtype = PaddleDType::FLOAT32; - std::vector outputs; - this->predictor_->Run({input_t}, &outputs, 1); - } + auto input_names = this->predictor_->GetInputNames(); + auto input_t = this->predictor_->GetInputHandle(input_names[0]); + input_t->Reshape({1, 3, resize_img.rows, resize_img.cols}); + input_t->CopyFromCpu(input.data()); + this->predictor_->Run(); std::vector out_data; auto output_names = this->predictor_->GetOutputNames(); - auto output_t = this->predictor_->GetOutputTensor(output_names[0]); + auto output_t = this->predictor_->GetOutputHandle(output_names[0]); std::vector output_shape = output_t->shape(); int out_num = std::accumulate(output_shape.begin(), output_shape.end(), 1, std::multiplies()); out_data.resize(out_num); - output_t->copy_to_cpu(out_data.data()); + output_t->CopyToCpu(out_data.data()); int n2 = output_shape[2]; int n3 = output_shape[3]; diff --git a/deploy/cpp_infer/src/ocr_rec.cpp b/deploy/cpp_infer/src/ocr_rec.cpp index d4deb5a1..e33695a7 100644 --- a/deploy/cpp_infer/src/ocr_rec.cpp +++ b/deploy/cpp_infer/src/ocr_rec.cpp @@ -43,32 +43,22 @@ void CRNNRecognizer::Run(std::vector>> boxes, this->permute_op_.Run(&resize_img, input.data()); // Inference. - if (this->use_zero_copy_run_) { - auto input_names = this->predictor_->GetInputNames(); - auto input_t = this->predictor_->GetInputTensor(input_names[0]); - input_t->Reshape({1, 3, resize_img.rows, resize_img.cols}); - input_t->copy_from_cpu(input.data()); - this->predictor_->ZeroCopyRun(); - } else { - paddle::PaddleTensor input_t; - input_t.shape = {1, 3, resize_img.rows, resize_img.cols}; - input_t.data = - paddle::PaddleBuf(input.data(), input.size() * sizeof(float)); - input_t.dtype = PaddleDType::FLOAT32; - std::vector outputs; - this->predictor_->Run({input_t}, &outputs, 1); - } + auto input_names = this->predictor_->GetInputNames(); + auto input_t = this->predictor_->GetInputHandle(input_names[0]); + input_t->Reshape({1, 3, resize_img.rows, resize_img.cols}); + input_t->CopyFromCpu(input.data()); + this->predictor_->Run(); std::vector predict_batch; auto output_names = this->predictor_->GetOutputNames(); - auto output_t = this->predictor_->GetOutputTensor(output_names[0]); + auto output_t = this->predictor_->GetOutputHandle(output_names[0]); auto predict_shape = output_t->shape(); int out_num = std::accumulate(predict_shape.begin(), predict_shape.end(), 1, std::multiplies()); predict_batch.resize(out_num); - output_t->copy_to_cpu(predict_batch.data()); + output_t->CopyToCpu(predict_batch.data()); // ctc decode std::vector str_res; @@ -102,7 +92,8 @@ void CRNNRecognizer::Run(std::vector>> boxes, } void CRNNRecognizer::LoadModel(const std::string &model_dir) { - AnalysisConfig config; + // AnalysisConfig config; + paddle_infer::Config config; config.SetModel(model_dir + "/inference.pdmodel", model_dir + "/inference.pdiparams"); @@ -118,9 +109,7 @@ void CRNNRecognizer::LoadModel(const std::string &model_dir) { config.SetCpuMathLibraryNumThreads(this->cpu_math_library_num_threads_); } - // false for zero copy tensor - // true for commom tensor - config.SwitchUseFeedFetchOps(!this->use_zero_copy_run_); + config.SwitchUseFeedFetchOps(false); // true for multiple input config.SwitchSpecifyInputNames(true); @@ -129,7 +118,7 @@ void CRNNRecognizer::LoadModel(const std::string &model_dir) { config.EnableMemoryOptim(); config.DisableGlogInfo(); - this->predictor_ = CreatePaddlePredictor(config); + this->predictor_ = CreatePredictor(config); } cv::Mat CRNNRecognizer::GetRotateCropImage(const cv::Mat &srcimage, diff --git a/deploy/cpp_infer/tools/config.txt b/deploy/cpp_infer/tools/config.txt index 95d7989b..87467aad 100644 --- a/deploy/cpp_infer/tools/config.txt +++ b/deploy/cpp_infer/tools/config.txt @@ -1,17 +1,16 @@ # model load config -use_gpu 0 +use_gpu 0 gpu_id 0 gpu_mem 4000 cpu_math_library_num_threads 10 use_mkldnn 0 -use_zero_copy_run 1 # det config max_side_len 960 det_db_thresh 0.3 det_db_box_thresh 0.5 det_db_unclip_ratio 2.0 -det_model_dir ./inference/ch__ppocr_mobile_v2.0_det_infer/ +det_model_dir ../../../deploy/cpp_infer/inference/ch_ppocr_mobile_v2.0_det_infer/ # cls config use_angle_cls 0 @@ -19,7 +18,7 @@ cls_model_dir ./inference/ch_ppocr_mobile_v2.0_cls_infer/ cls_thresh 0.9 # rec config -rec_model_dir ./inference/ch_ppocr_mobile_v2.0_rec_infer/ +rec_model_dir ../../../deploy/cpp_infer/inference/ch_ppocr_mobile_v2.0_rec_infer/ char_list_file ../../ppocr/utils/ppocr_keys_v1.txt # show the detection results From 885378fd33f659f387aabc86c2b3de065db06d84 Mon Sep 17 00:00:00 2001 From: LDOUBLEV Date: Mon, 21 Dec 2020 13:45:02 +0000 Subject: [PATCH 40/60] fix config.txt --- deploy/cpp_infer/tools/config.txt | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/deploy/cpp_infer/tools/config.txt b/deploy/cpp_infer/tools/config.txt index 3aa6d582..34f47ed8 100644 --- a/deploy/cpp_infer/tools/config.txt +++ b/deploy/cpp_infer/tools/config.txt @@ -18,7 +18,7 @@ cls_model_dir ./inference/ch_ppocr_mobile_v2.0_cls_infer/ cls_thresh 0.9 # rec config -rec_model_dir ../../../deploy/cpp_infer/inference/ch_ppocr_mobile_v2.0_rec_infer/ +rec_model_dir ./inference/ch_ppocr_mobile_v2.0_rec_infer/ char_list_file ../../ppocr/utils/ppocr_keys_v1.txt # show the detection results From d49aff19618eead1131feb5c1092ebe498a92df1 Mon Sep 17 00:00:00 2001 From: LDOUBLEV Date: Mon, 21 Dec 2020 13:54:37 +0000 Subject: [PATCH 41/60] fix mkl warning --- deploy/cpp_infer/src/main.cpp | 13 ------------- 1 file changed, 13 deletions(-) diff --git a/deploy/cpp_infer/src/main.cpp b/deploy/cpp_infer/src/main.cpp index 187b70dc..21890d45 100644 --- a/deploy/cpp_infer/src/main.cpp +++ b/deploy/cpp_infer/src/main.cpp @@ -67,19 +67,6 @@ int main(int argc, char **argv) { config.gpu_mem, config.cpu_math_library_num_threads, config.use_mkldnn, config.char_list_file); -#ifdef USE_MKL -#pragma omp parallel - for (auto i = 0; i < 10; i++) { - LOG_IF(WARNING, - config.cpu_math_library_num_threads != omp_get_num_threads()) - << "WARNING! MKL is running on " << omp_get_num_threads() - << " threads while cpu_math_library_num_threads is set to " - << config.cpu_math_library_num_threads - << ". Possible reason could be 1. You have set omp_set_num_threads() " - "somewhere; 2. MKL is not linked properly"; - } -#endif - auto start = std::chrono::system_clock::now(); std::vector>> boxes; det.Run(srcimg, boxes); From d3bd5103772fe84ce15e3602bce72f1c595702fb Mon Sep 17 00:00:00 2001 From: zhoujun <572459439@qq.com> Date: Mon, 21 Dec 2020 22:27:11 +0800 Subject: [PATCH 42/60] fix bugs in DetResizeForTest --- ppocr/data/imaug/operators.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/ppocr/data/imaug/operators.py b/ppocr/data/imaug/operators.py index 57cd3b4b..8b9175cf 100644 --- a/ppocr/data/imaug/operators.py +++ b/ppocr/data/imaug/operators.py @@ -119,10 +119,10 @@ class DetResizeForTest(object): if 'image_shape' in kwargs: self.image_shape = kwargs['image_shape'] self.resize_type = 1 - if 'limit_side_len' in kwargs: + elif 'limit_side_len' in kwargs: self.limit_side_len = kwargs['limit_side_len'] self.limit_type = kwargs.get('limit_type', 'min') - if 'resize_long' in kwargs: + elif 'resize_long' in kwargs: self.resize_type = 2 self.resize_long = kwargs.get('resize_long', 960) else: From dbd27878cb59789b52affff3556ee070d2c65cfa Mon Sep 17 00:00:00 2001 From: LDOUBLEV Date: Tue, 22 Dec 2020 10:09:51 +0800 Subject: [PATCH 43/60] update cpp_infer readme --- deploy/cpp_infer/readme.md | 15 +++++++-------- deploy/cpp_infer/readme_en.md | 19 ++++++++++--------- 2 files changed, 17 insertions(+), 17 deletions(-) diff --git a/deploy/cpp_infer/readme.md b/deploy/cpp_infer/readme.md index 66302a01..b563ecf4 100644 --- a/deploy/cpp_infer/readme.md +++ b/deploy/cpp_infer/readme.md @@ -122,10 +122,10 @@ build/paddle_inference_install_dir/ * 下载之后使用下面的方法解压。 ``` -tar -xf fluid_inference.tgz +tar -xf paddle_inference.tgz ``` -最终会在当前的文件夹中生成`fluid_inference/`的子文件夹。 +最终会在当前的文件夹中生成`paddle_inference/`的子文件夹。 ## 2 开始运行 @@ -137,11 +137,11 @@ tar -xf fluid_inference.tgz ``` inference/ |-- det_db -| |--model -| |--params +| |--inference.pdparams +| |--inference.pdimodel |-- rec_rcnn -| |--model -| |--params +| |--inference.pdparams +| |--inference.pdparams ``` @@ -180,7 +180,7 @@ cmake .. \ make -j ``` -`OPENCV_DIR`为opencv编译安装的地址；`LIB_DIR`为下载(`fluid_inference`文件夹)或者编译生成的Paddle预测库地址(`build/fluid_inference_install_dir`文件夹)；`CUDA_LIB_DIR`为cuda库文件地址，在docker中；为`/usr/local/cuda/lib64`；`CUDNN_LIB_DIR`为cudnn库文件地址，在docker中为`/usr/lib/x86_64-linux-gnu/`。 +`OPENCV_DIR`为opencv编译安装的地址；`LIB_DIR`为下载(`paddle_inference`文件夹)或者编译生成的Paddle预测库地址(`build/paddle_inference_install_dir`文件夹)；`CUDA_LIB_DIR`为cuda库文件地址，在docker中；为`/usr/local/cuda/lib64`；`CUDNN_LIB_DIR`为cudnn库文件地址，在docker中为`/usr/lib/x86_64-linux-gnu/`。 * 编译完成之后，会在`build`文件夹下生成一个名为`ocr_system`的可执行文件。 @@ -202,7 +202,6 @@ gpu_id 0 # GPU id，使用GPU时有效 gpu_mem 4000 # 申请的GPU内存 cpu_math_library_num_threads 10 # CPU预测时的线程数，在机器核数充足的情况下，该值越大，预测速度越快 use_mkldnn 1 # 是否使用mkldnn库 -use_zero_copy_run 1 # 是否使用use_zero_copy_run进行预测 # det config max_side_len 960 # 输入图像长宽大于960时，等比例缩放图像，使得图像最长边为960 diff --git a/deploy/cpp_infer/readme_en.md b/deploy/cpp_infer/readme_en.md index 8bd76c04..41c764bc 100644 --- a/deploy/cpp_infer/readme_en.md +++ b/deploy/cpp_infer/readme_en.md @@ -130,10 +130,10 @@ Among them, `paddle` is the Paddle library required for C++ prediction later, an * After downloading, use the following method to uncompress. ``` -tar -xf fluid_inference.tgz +tar -xf paddle_inference.tgz ``` -Finally you can see the following files in the folder of `fluid_inference/`. +Finally you can see the following files in the folder of `paddle_inference/`. ## 2. Compile and run the demo @@ -145,11 +145,11 @@ Finally you can see the following files in the folder of `fluid_inference/`. ``` inference/ |-- det_db -| |--model -| |--params +| |--inference.pdparams +| |--inference.pdimodel |-- rec_rcnn -| |--model -| |--params +| |--inference.pdparams +| |--inference.pdparams ``` @@ -188,7 +188,9 @@ cmake .. \ make -j ``` -`OPENCV_DIR` is the opencv installation path; `LIB_DIR` is the download (`fluid_inference` folder) or the generated Paddle inference library path (`build/fluid_inference_install_dir` folder); `CUDA_LIB_DIR` is the cuda library file path, in docker; it is `/usr/local/cuda/lib64`; `CUDNN_LIB_DIR` is the cudnn library file path, in docker it is `/usr/lib/x86_64-linux-gnu/`. +`OPENCV_DIR` is the opencv installation path; `LIB_DIR` is the download (`paddle_inference` folder) +or the generated Paddle inference library path (`build/paddle_inference_install_dir` folder); +`CUDA_LIB_DIR` is the cuda library file path, in docker; it is `/usr/local/cuda/lib64`; `CUDNN_LIB_DIR` is the cudnn library file path, in docker it is `/usr/lib/x86_64-linux-gnu/`. * After the compilation is completed, an executable file named `ocr_system` will be generated in the `build` folder. @@ -211,7 +213,6 @@ gpu_id 0 # GPU id when use_gpu is 1 gpu_mem 4000 # GPU memory requested cpu_math_library_num_threads 10 # Number of threads when using CPU inference. When machine cores is enough, the large the value, the faster the inference speed use_mkldnn 1 # Whether to use mkdlnn library -use_zero_copy_run 1 # Whether to use use_zero_copy_run for inference max_side_len 960 # Limit the maximum image height and width to 960 det_db_thresh 0.3 # Used to filter the binarized image of DB prediction, setting 0.-0.3 has no obvious effect on the result @@ -244,4 +245,4 @@ The detection results will be shown on the screen, which is as follows. ### 2.3 Notes -* Paddle2.0.0-beta0 inference model library is recommanded for this tuturial. +* Paddle2.0.0-beta0 inference model library is recommended for this toturial. From 2a04e1a12a422941b55e794df9d49fcc312b2e0b Mon Sep 17 00:00:00 2001 From: LDOUBLEV Date: Tue, 22 Dec 2020 15:57:21 +0800 Subject: [PATCH 44/60] add export to limit GPU-memory-usage --- tools/infer/predict_cls.py | 2 ++ tools/infer/predict_det.py | 2 ++ tools/infer/predict_rec.py | 2 ++ tools/infer/predict_system.py | 2 ++ tools/infer/utility.py | 5 +++-- tools/infer_cls.py | 2 ++ tools/infer_det.py | 2 ++ tools/infer_rec.py | 2 ++ 8 files changed, 17 insertions(+), 2 deletions(-) diff --git a/tools/infer/predict_cls.py b/tools/infer/predict_cls.py index 6604d3cc..074172cc 100755 --- a/tools/infer/predict_cls.py +++ b/tools/infer/predict_cls.py @@ -18,6 +18,8 @@ __dir__ = os.path.dirname(os.path.abspath(__file__)) sys.path.append(__dir__) sys.path.append(os.path.abspath(os.path.join(__dir__, '../..'))) +os.environ["FLAGS_allocator_strategy"] = 'auto_growth' + import cv2 import copy import numpy as np diff --git a/tools/infer/predict_det.py b/tools/infer/predict_det.py index e677eb24..077692af 100755 --- a/tools/infer/predict_det.py +++ b/tools/infer/predict_det.py @@ -18,6 +18,8 @@ __dir__ = os.path.dirname(os.path.abspath(__file__)) sys.path.append(__dir__) sys.path.append(os.path.abspath(os.path.join(__dir__, '../..'))) +os.environ["FLAGS_allocator_strategy"] = 'auto_growth' + import cv2 import numpy as np import time diff --git a/tools/infer/predict_rec.py b/tools/infer/predict_rec.py index 89d04cff..974fdbb6 100755 --- a/tools/infer/predict_rec.py +++ b/tools/infer/predict_rec.py @@ -18,6 +18,8 @@ __dir__ = os.path.dirname(os.path.abspath(__file__)) sys.path.append(__dir__) sys.path.append(os.path.abspath(os.path.join(__dir__, '../..'))) +os.environ["FLAGS_allocator_strategy"] = 'auto_growth' + import cv2 import numpy as np import math diff --git a/tools/infer/predict_system.py b/tools/infer/predict_system.py index 07dfc216..8c4f9214 100755 --- a/tools/infer/predict_system.py +++ b/tools/infer/predict_system.py @@ -18,6 +18,8 @@ __dir__ = os.path.dirname(os.path.abspath(__file__)) sys.path.append(__dir__) sys.path.append(os.path.abspath(os.path.join(__dir__, '../..'))) +os.environ["FLAGS_allocator_strategy"] = 'auto_growth' + import cv2 import copy import numpy as np diff --git a/tools/infer/utility.py b/tools/infer/utility.py index 39c045de..966fa3cc 100755 --- a/tools/infer/utility.py +++ b/tools/infer/utility.py @@ -33,7 +33,7 @@ def parse_args(): parser.add_argument("--ir_optim", type=str2bool, default=True) parser.add_argument("--use_tensorrt", type=str2bool, default=False) parser.add_argument("--use_fp16", type=str2bool, default=False) - parser.add_argument("--gpu_mem", type=int, default=8000) + parser.add_argument("--gpu_mem", type=int, default=500) # params for text detector parser.add_argument("--image_dir", type=str) @@ -62,7 +62,7 @@ def parse_args(): parser.add_argument("--rec_model_dir", type=str) parser.add_argument("--rec_image_shape", type=str, default="3, 32, 320") parser.add_argument("--rec_char_type", type=str, default='ch') - parser.add_argument("--rec_batch_num", type=int, default=1) + parser.add_argument("--rec_batch_num", type=int, default=6) parser.add_argument("--max_text_length", type=int, default=25) parser.add_argument( "--rec_char_dict_path", @@ -123,6 +123,7 @@ def create_predictor(args, mode, logger): # cache 10 different shapes for mkldnn to avoid memory leak config.set_mkldnn_cache_capacity(10) config.enable_mkldnn() + args.rec_batch_num = 1 # config.enable_memory_optim() config.disable_glog_info() diff --git a/tools/infer_cls.py b/tools/infer_cls.py index 85e11ac3..49696482 100755 --- a/tools/infer_cls.py +++ b/tools/infer_cls.py @@ -25,6 +25,8 @@ __dir__ = os.path.dirname(os.path.abspath(__file__)) sys.path.append(__dir__) sys.path.append(os.path.abspath(os.path.join(__dir__, '..'))) +os.environ["FLAGS_allocator_strategy"] = 'auto_growth' + import paddle from ppocr.data import create_operators, transform diff --git a/tools/infer_det.py b/tools/infer_det.py index d1b1b752..d890970e 100755 --- a/tools/infer_det.py +++ b/tools/infer_det.py @@ -25,6 +25,8 @@ __dir__ = os.path.dirname(os.path.abspath(__file__)) sys.path.append(__dir__) sys.path.append(os.path.abspath(os.path.join(__dir__, '..'))) +os.environ["FLAGS_allocator_strategy"] = 'auto_growth' + import cv2 import json import paddle diff --git a/tools/infer_rec.py b/tools/infer_rec.py index e3e85b5d..7e4b0811 100755 --- a/tools/infer_rec.py +++ b/tools/infer_rec.py @@ -25,6 +25,8 @@ __dir__ = os.path.dirname(os.path.abspath(__file__)) sys.path.append(__dir__) sys.path.append(os.path.abspath(os.path.join(__dir__, '..'))) +os.environ["FLAGS_allocator_strategy"] = 'auto_growth' + import paddle from ppocr.data import create_operators, transform From 091574bbd1cc844ff7bfefbc51d9730e7a8733c5 Mon Sep 17 00:00:00 2001 From: WenmuZhou Date: Thu, 24 Dec 2020 17:17:28 +0800 Subject: [PATCH 45/60] fix bug --- configs/det/det_r50_vd_sast_totaltext.yml | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/configs/det/det_r50_vd_sast_totaltext.yml b/configs/det/det_r50_vd_sast_totaltext.yml index a92f1b6e..e040c420 100755 --- a/configs/det/det_r50_vd_sast_totaltext.yml +++ b/configs/det/det_r50_vd_sast_totaltext.yml @@ -62,7 +62,7 @@ Train: name: SimpleDataSet data_dir: ./train_data/ label_file_list: [./train_data/art_latin_icdar_14pt/train_no_tt_test/train_label_json.txt, ./train_data/total_text_icdar_14pt/train_label_json.txt] - data_ratio_list: [0.5, 0.5] + ratio_list: [0.5, 0.5] transforms: - DecodeImage: # load image img_mode: BGR From 31aa45fed220e76feaaff717343e4e67f23e14a7 Mon Sep 17 00:00:00 2001 From: WenmuZhou Date: Fri, 25 Dec 2020 16:39:50 +0800 Subject: [PATCH 46/60] add init to fc1 --- ppocr/modeling/transforms/tps.py | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/ppocr/modeling/transforms/tps.py b/ppocr/modeling/transforms/tps.py index 86665bed..e7a152c1 100644 --- a/ppocr/modeling/transforms/tps.py +++ b/ppocr/modeling/transforms/tps.py @@ -16,6 +16,7 @@ from __future__ import absolute_import from __future__ import division from __future__ import print_function +import math import paddle from paddle import nn, ParamAttr from paddle.nn import functional as F @@ -88,11 +89,14 @@ class LocalizationNetwork(nn.Layer): in_channels = num_filters self.block_list.append(pool) name = "loc_fc1" + stdv = 1.0 / math.sqrt(num_filters_list[-1] * 1.0) self.fc1 = nn.Linear( in_channels, fc_dim, weight_attr=ParamAttr( - learning_rate=loc_lr, name=name + "_w"), + learning_rate=loc_lr, + name=name + "_w", + initializer=nn.initializer.Uniform(-stdv, stdv)), bias_attr=ParamAttr(name=name + '.b_0'), name=name) From 2a2c891d6bfd7f0887b0b9cfd09dbac6ece82909 Mon Sep 17 00:00:00 2001 From: zhoujun <572459439@qq.com> Date: Sat, 26 Dec 2020 23:40:24 +0800 Subject: [PATCH 47/60] update ratio_list in db config --- configs/det/det_mv3_db.yml | 2 +- configs/det/det_r50_vd_db.yml | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/configs/det/det_mv3_db.yml b/configs/det/det_mv3_db.yml index 5c8a0923..bdb4afc0 100644 --- a/configs/det/det_mv3_db.yml +++ b/configs/det/det_mv3_db.yml @@ -67,7 +67,7 @@ Train: data_dir: ./train_data/icdar2015/text_localization/ label_file_list: - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt - ratio_list: [0.5] + ratio_list: [1.0] transforms: - DecodeImage: # load image img_mode: BGR diff --git a/configs/det/det_r50_vd_db.yml b/configs/det/det_r50_vd_db.yml index f1188fe3..19c059d6 100644 --- a/configs/det/det_r50_vd_db.yml +++ b/configs/det/det_r50_vd_db.yml @@ -66,7 +66,7 @@ Train: data_dir: ./train_data/icdar2015/text_localization/ label_file_list: - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt - ratio_list: [0.5] + ratio_list: [1.0] transforms: - DecodeImage: # load image img_mode: BGR From 8985f6c207fa36f52eb562f68afcc26c4dee149c Mon Sep 17 00:00:00 2001 From: littletomatodonkey <2120160898@bit.edu.cn> Date: Mon, 28 Dec 2020 14:58:32 +0800 Subject: [PATCH 48/60] add faq 20201218 (#1590) --- README_ch.md | 4 +-- doc/doc_ch/FAQ.md | 70 +++++++++++++++++++++++++++-------------------- 2 files changed, 42 insertions(+), 32 deletions(-) diff --git a/README_ch.md b/README_ch.md index 3954bf45..3b0a2bfb 100755 --- a/README_ch.md +++ b/README_ch.md @@ -8,7 +8,7 @@ PaddleOCR同时支持动态图与静态图两种编程范式 - 静态图版本：develop分支 **近期更新** -- 2020.12.21 [FAQ](./doc/doc_ch/FAQ.md)新增5个高频问题，总数132个，每周一都会更新，欢迎大家持续关注。 +- 2020.12.28 [FAQ](./doc/doc_ch/FAQ.md)新增5个高频问题，总数137个，每周一都会更新，欢迎大家持续关注。 - 2020.12.15 更新数据合成工具[Style-Text](./StyleText/README_ch.md)，可以批量合成大量与目标场景类似的图像，在多个场景验证，效果明显提升。 - 2020.11.25 更新半自动标注工具[PPOCRLabel](./PPOCRLabel/README_ch.md)，辅助开发者高效完成标注任务，输出格式与PP-OCR训练任务完美衔接。 - 2020.9.22 更新PP-OCR技术文章，https://arxiv.org/abs/2009.09941 @@ -102,7 +102,7 @@ PaddleOCR同时支持动态图与静态图两种编程范式 - FAQ - [【精选】OCR精选10个问题](./doc/doc_ch/FAQ.md) - [【理论篇】OCR通用31个问题](./doc/doc_ch/FAQ.md) - - [【实战篇】PaddleOCR实战91个问题](./doc/doc_ch/FAQ.md) + - [【实战篇】PaddleOCR实战96个问题](./doc/doc_ch/FAQ.md) - [技术交流群](#欢迎加入PaddleOCR技术交流群) - [参考文献](./doc/doc_ch/reference.md) - [许可证书](#许可证书) diff --git a/doc/doc_ch/FAQ.md b/doc/doc_ch/FAQ.md index 288a5492..1abf30cc 100755 --- a/doc/doc_ch/FAQ.md +++ b/doc/doc_ch/FAQ.md @@ -9,54 +9,44 @@ ## PaddleOCR常见问题汇总(持续更新) -* [近期更新（2020.12.21）](#近期更新) +* [近期更新（2020.12.28）](#近期更新) * [【精选】OCR精选10个问题](#OCR精选10个问题) * [【理论篇】OCR通用31个问题](#OCR通用问题) * [基础知识7题](#基础知识) * [数据集7题](#数据集2) * [模型训练调优17题](#模型训练调优2) -* [【实战篇】PaddleOCR实战91个问题](#PaddleOCR实战问题) - * [使用咨询24题](#使用咨询) +* [【实战篇】PaddleOCR实战96个问题](#PaddleOCR实战问题) + * [使用咨询28题](#使用咨询) * [数据集17题](#数据集3) - * [模型训练调优25题](#模型训练调优3) + * [模型训练调优26题](#模型训练调优3) * [预测部署25题](#预测部署3) -## 近期更新（2020.12.21） +## 近期更新（2020.12.28） -#### Q2.3.17: StyleText 合成数据效果不好？ -**A**： StyleText模型生成的数据主要用于OCR识别模型的训练。PaddleOCR目前识别模型的输入为32 x N，因此当前版本模型主要适用高度为32的数据。 -建议要合成的数据尺寸设置为32 x N。尺寸相差不多的数据也可以生成，尺寸很大或很小的数据效果确实不佳。 +#### Q3.1.25: 使用dygraph分支，在docker中训练PaddleOCR的时候，数据路径没有任何问题，但是一直报错`reader rasied an exception`，这是为什么呢？ -#### Q3.1.22： ModuleNotFoundError: No module named 'paddle.nn'， -**A**： paddle.nn是Paddle2.0版本特有的功能，请安装大于等于Paddle 2.0.0rc1的版本，安装方式为 -``` -python3 -m pip install paddlepaddle-gpu==2.0.0rc1 -i https://mirror.baidu.com/pypi/simple -``` +**A** 创建docker的时候，`/dev/shm`的默认大小为64M，如果使用多进程读取数据，共享内存可能不够，因此需要给`/dev/shm`分配更大的空间，在创建docker的时候，传入`--shm-size=8g`表示给`/dev/shm`分配8g的空间。 -#### Q3.1.23： ImportError: /usr/lib/x86_64_linux-gnu/libstdc++.so.6:version `CXXABI_1.3.11` not found (required by /usr/lib/python3.6/site-package/paddle/fluid/core+avx.so) -**A**：这个问题是glibc版本不足导致的，Paddle2.0rc1版本对gcc版本和glib版本有更高的要求，推荐gcc版本为8.2，glibc版本2.12以上。 -如果您的环境不满足这个要求，或者使用的docker镜像为: -`hub.baidubce.com/paddlepaddle/paddle:latest-gpu-cuda9.0-cudnn7-dev` -`hub.baidubce.com/paddlepaddle/paddle:latest-gpu-cuda9.0-cudnn7-dev`，安装Paddle2.0rc版本可能会出现上述错误， -2.0版本推荐使用新的docker镜像 `paddlepaddle/paddle:latest-dev-cuda10.1-cudnn7-gcc82`。 -或者访问[dockerhub](https://hub.docker.com/r/paddlepaddle/paddle/tags/)获得与您机器适配的镜像。 +#### Q3.1.26: 在repo中没有找到Lite和PaddleServing相关的部署教程，这是在哪里呢？ -#### Q3.1.24: PaddleOCR develop分支和dygraph分支的区别？ -**A**：目前PaddleOCR有四个分支，分别是： -- develop：基于Paddle静态图开发的分支，推荐使用paddle1.8 或者2.0版本，该分支具备完善的模型训练、预测、推理部署、量化裁剪等功能，领先于release/1.1分支。 -- release/1.1：PaddleOCR 发布的第一个稳定版本，基于静态图开发，具备完善的训练、预测、推理部署、量化裁剪等功能。 -- dygraph：基于Paddle动态图开发的分支，未来将作为主要开发分支，运行要求使用Paddle2.0rc1版本，目前仍在开发中。 -- release/2.0-rc1-0：PaddleOCR发布的第二个稳定版本，基于动态图和paddle2.0版本开发，动态图开发的工程更易于调试，目前支，支持模型训练、预测，暂不支持移动端部署。 +**A** 目前PaddleOCR的默认分支为dygraph，关于Lite和PaddleLite的动态图部署还在适配中，如果希望在Lite端或者使用PaddleServing部署，推荐使用develop分支（静态图）的代码。 -如果您已经上手过PaddleOCR，并且希望在各种环境上部署PaddleOCR，目前建议使用静态图分支，develop或者release/1.1分支。如果您是初学者，想快速训练，调试PaddleOCR中的算法，建议尝鲜PaddleOCR dygraph分支。 +#### Q3.1.27: 如何可视化acc,loss曲线图,模型网络结构图等？ -**注意**：develop和dygraph分支要求的Paddle版本、本地环境有差别，请注意不同分支环境安装部分的差异。 +**A** 在配置文件里有`use_visualdl`的参数，设置为True即可，更多的使用命令可以参考：[VisualDL使用指南](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0-rc1/guides/03_VisualDL/visualdl.html)。 +#### Q3.1.28: 在使用StyleText数据合成工具的时候，报错`ModuleNotFoundError: No module named 'utils.config'`，这是为什么呢？ + +**A** 有2个解决方案 +- 在StyleText路径下面设置PYTHONPATH：`export PYTHONPATH=./` +- 拉取最新的代码 + +#### Q3.3.26: PaddleOCR在训练的时候一直使用cosine_decay的学习率下降策略，这是为什么呢？ + +**A**: cosine_decay表示在训练的过程中，学习率按照cosine的变化趋势逐渐下降至0，在迭代轮数更长的情况下，比常量的学习率变化策略会有更好的收敛效果，因此在实际训练的时候，均采用了cosine_decay，来获得精度更高的模型。 -#### Q3.4.25 : PaddleOCR模型Python端预测和C++预测结果不一致？ -**A**：正常来说，python端预测和C++预测文本是一致的，如果预测结果差异较大，建议首先排查diff出现在检测模型还是识别模型，或者尝试换其他模型是否有类似的问题。其次，检查python端和C++端数据处理部分是否存在差异，建议保存环境，更新PaddleOCR代码再试下。如果更新代码或者更新代码都没能解决，建议在PaddleOCR群里或者issue中抛出您的问题。 @@ -431,7 +421,23 @@ python3 -m pip install paddlepaddle-gpu==2.0.0rc1 -i https://mirror.baidu.com/py **注意**：develop和dygraph分支要求的Paddle版本、本地环境有差别，请注意不同分支环境安装部分的差异。 +#### Q3.1.25: 使用dygraph分支，在docker中训练PaddleOCR的时候，数据路径没有任何问题，但是一直报错`reader rasied an exception`，这是为什么呢？ +**A** 创建docker的时候，`/dev/shm`的默认大小为64M，如果使用多进程读取数据，共享内存可能不够，因此需要给`/dev/shm`分配更大的空间，在创建docker的时候，传入`--shm-size=8g`表示给`/dev/shm`分配8g的空间。 + +#### Q3.1.26: 在repo中没有找到Lite和PaddleServing相关的部署教程，这是在哪里呢？ + +**A** 目前PaddleOCR的默认分支为dygraph，关于Lite和PaddleLite的动态图部署还在适配中，如果希望在Lite端或者使用PaddleServing部署，推荐使用develop分支（静态图）的代码。 + +#### Q3.1.27: 如何可视化acc,loss曲线图,模型网络结构图等？ + +**A** 在配置文件里有`use_visualdl`的参数，设置为True即可，更多的使用命令可以参考：[VisualDL使用指南](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0-rc1/guides/03_VisualDL/visualdl.html)。 + +#### Q3.1.28: 在使用StyleText数据合成工具的时候，报错`ModuleNotFoundError: No module named 'utils.config'`，这是为什么呢？ + +**A** 有2个解决方案 +- 在StyleText路径下面设置PYTHONPATH：`export PYTHONPATH=./` +- 拉取最新的代码 ### 数据集 @@ -658,6 +664,10 @@ ps -axu | grep train.py | awk '{print $2}' | xargs kill -9 **A**: 识别模型训练初期acc为0是正常的，多训一段时间指标就上来了。 +#### Q3.3.26: PaddleOCR在训练的时候一直使用cosine_decay的学习率下降策略，这是为什么呢？ + +**A**: cosine_decay表示在训练的过程中，学习率按照cosine的变化趋势逐渐下降至0，在迭代轮数更长的情况下，比常量的学习率变化策略会有更好的收敛效果，因此在实际训练的时候，均采用了cosine_decay，来获得精度更高的模型。 + ### 预测部署 From de3e2e7cd3b8b65ee02d7a41e570fa5b511a3c1d Mon Sep 17 00:00:00 2001 From: littletomatodonkey <2120160898@bit.edu.cn> Date: Tue, 29 Dec 2020 13:49:43 +0800 Subject: [PATCH 49/60] add CyclicalCosineDecay (#1599) --- ppocr/optimizer/learning_rate.py | 51 +++++++++++++++++++++++++++++--- ppocr/optimizer/lr_scheduler.py | 49 ++++++++++++++++++++++++++++++ tools/program.py | 4 +-- 3 files changed, 98 insertions(+), 6 deletions(-) create mode 100644 ppocr/optimizer/lr_scheduler.py diff --git a/ppocr/optimizer/learning_rate.py b/ppocr/optimizer/learning_rate.py index 8f303e83..e1b10992 100644 --- a/ppocr/optimizer/learning_rate.py +++ b/ppocr/optimizer/learning_rate.py @@ -18,6 +18,7 @@ from __future__ import print_function from __future__ import unicode_literals from paddle.optimizer import lr +from .lr_scheduler import CyclicalCosineDecay class Linear(object): @@ -46,7 +47,7 @@ class Linear(object): self.end_lr = end_lr self.power = power self.last_epoch = last_epoch - self.warmup_epoch = warmup_epoch * step_each_epoch + self.warmup_epoch = round(warmup_epoch * step_each_epoch) def __call__(self): learning_rate = lr.PolynomialDecay( @@ -87,7 +88,7 @@ class Cosine(object): self.learning_rate = learning_rate self.T_max = step_each_epoch * epochs self.last_epoch = last_epoch - self.warmup_epoch = warmup_epoch * step_each_epoch + self.warmup_epoch = round(warmup_epoch * step_each_epoch) def __call__(self): learning_rate = lr.CosineAnnealingDecay( @@ -129,7 +130,7 @@ class Step(object): self.learning_rate = learning_rate self.gamma = gamma self.last_epoch = last_epoch - self.warmup_epoch = warmup_epoch * step_each_epoch + self.warmup_epoch = round(warmup_epoch * step_each_epoch) def __call__(self): learning_rate = lr.StepDecay( @@ -168,7 +169,7 @@ class Piecewise(object): self.boundaries = [step_each_epoch * e for e in decay_epochs] self.values = values self.last_epoch = last_epoch - self.warmup_epoch = warmup_epoch * step_each_epoch + self.warmup_epoch = round(warmup_epoch * step_each_epoch) def __call__(self): learning_rate = lr.PiecewiseDecay( @@ -183,3 +184,45 @@ class Piecewise(object): end_lr=self.values[0], last_epoch=self.last_epoch) return learning_rate + + +class CyclicalCosine(object): + """ + Cyclical cosine learning rate decay + Args: + learning_rate(float): initial learning rate + step_each_epoch(int): steps each epoch + epochs(int): total training epochs + cycle(int): period of the cosine learning rate + last_epoch (int, optional): The index of last epoch. Can be set to restart training. Default: -1, means initial learning rate. + """ + + def __init__(self, + learning_rate, + step_each_epoch, + epochs, + cycle, + warmup_epoch=0, + last_epoch=-1, + **kwargs): + super(CyclicalCosine, self).__init__() + self.learning_rate = learning_rate + self.T_max = step_each_epoch * epochs + self.last_epoch = last_epoch + self.warmup_epoch = round(warmup_epoch * step_each_epoch) + self.cycle = round(cycle * step_each_epoch) + + def __call__(self): + learning_rate = CyclicalCosineDecay( + learning_rate=self.learning_rate, + T_max=self.T_max, + cycle=self.cycle, + last_epoch=self.last_epoch) + if self.warmup_epoch > 0: + learning_rate = lr.LinearWarmup( + learning_rate=learning_rate, + warmup_steps=self.warmup_epoch, + start_lr=0.0, + end_lr=self.learning_rate, + last_epoch=self.last_epoch) + return learning_rate diff --git a/ppocr/optimizer/lr_scheduler.py b/ppocr/optimizer/lr_scheduler.py new file mode 100644 index 00000000..21aec737 --- /dev/null +++ b/ppocr/optimizer/lr_scheduler.py @@ -0,0 +1,49 @@ +# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import math +from paddle.optimizer.lr import LRScheduler + + +class CyclicalCosineDecay(LRScheduler): + def __init__(self, + learning_rate, + T_max, + cycle=1, + last_epoch=-1, + eta_min=0.0, + verbose=False): + """ + Cyclical cosine learning rate decay + A learning rate which can be referred in https://arxiv.org/pdf/2012.12645.pdf + Args: + learning rate(float): learning rate + T_max(int): maximum epoch num + cycle(int): period of the cosine decay + last_epoch (int, optional): The index of last epoch. Can be set to restart training. Default: -1, means initial learning rate. + eta_min(float): minimum learning rate during training + verbose(bool): whether to print learning rate for each epoch + """ + super(CyclicalCosineDecay, self).__init__(learning_rate, last_epoch, + verbose) + self.cycle = cycle + self.eta_min = eta_min + + def get_lr(self): + if self.last_epoch == 0: + return self.base_lr + reletive_epoch = self.last_epoch % self.cycle + lr = self.eta_min + 0.5 * (self.base_lr - self.eta_min) * \ + (1 + math.cos(math.pi * reletive_epoch / self.cycle)) + return lr diff --git a/tools/program.py b/tools/program.py index c712fe14..c2915426 100755 --- a/tools/program.py +++ b/tools/program.py @@ -179,9 +179,9 @@ def train(config, if 'start_epoch' in best_model_dict: start_epoch = best_model_dict['start_epoch'] else: - start_epoch = 0 + start_epoch = 1 - for epoch in range(start_epoch, epoch_num): + for epoch in range(start_epoch, epoch_num + 1): if epoch > 0: train_dataloader = build_dataloader(config, 'Train', device, logger) train_batch_cost = 0.0 From c03b0b0604f238f918b30dff3c043fb467ea27f2 Mon Sep 17 00:00:00 2001 From: zhoujun Date: Thu, 31 Dec 2020 16:31:21 +0800 Subject: [PATCH 50/60] Update utility.py --- tools/infer/utility.py | 1 + 1 file changed, 1 insertion(+) diff --git a/tools/infer/utility.py b/tools/infer/utility.py index 966fa3cc..c19946eb 100755 --- a/tools/infer/utility.py +++ b/tools/infer/utility.py @@ -83,6 +83,7 @@ def parse_args(): parser.add_argument("--enable_mkldnn", type=str2bool, default=False) parser.add_argument("--use_pdserving", type=str2bool, default=False) + parser.add_argument("--use_tensorrt", type=str2bool, default=False) return parser.parse_args() From 499ae66003c278cba948508af812adc80033496d Mon Sep 17 00:00:00 2001 From: zhoujun <572459439@qq.com> Date: Fri, 1 Jan 2021 21:53:15 +0800 Subject: [PATCH 51/60] update utility --- tools/infer/utility.py | 1 - 1 file changed, 1 deletion(-) diff --git a/tools/infer/utility.py b/tools/infer/utility.py index c19946eb..966fa3cc 100755 --- a/tools/infer/utility.py +++ b/tools/infer/utility.py @@ -83,7 +83,6 @@ def parse_args(): parser.add_argument("--enable_mkldnn", type=str2bool, default=False) parser.add_argument("--use_pdserving", type=str2bool, default=False) - parser.add_argument("--use_tensorrt", type=str2bool, default=False) return parser.parse_args() From c465f40417d24b9bd791b8c05344f1b5f52548fa Mon Sep 17 00:00:00 2001 From: WenmuZhou Date: Mon, 4 Jan 2021 12:11:21 +0800 Subject: [PATCH 52/60] add tensorrt to hubserving --- deploy/hubserving/ocr_cls/params.py | 2 +- deploy/hubserving/ocr_det/params.py | 2 +- deploy/hubserving/ocr_rec/params.py | 4 ++-- deploy/hubserving/ocr_system/params.py | 2 +- 4 files changed, 5 insertions(+), 5 deletions(-) diff --git a/deploy/hubserving/ocr_cls/params.py b/deploy/hubserving/ocr_cls/params.py index 72a7a102..982f0136 100755 --- a/deploy/hubserving/ocr_cls/params.py +++ b/deploy/hubserving/ocr_cls/params.py @@ -18,7 +18,7 @@ def read_params(): cfg.cls_batch_num = 30 cfg.cls_thresh = 0.9 - cfg.use_zero_copy_run = False cfg.use_pdserving = False + cfg.use_tensorrt = False return cfg diff --git a/deploy/hubserving/ocr_det/params.py b/deploy/hubserving/ocr_det/params.py index e50decbb..13215890 100755 --- a/deploy/hubserving/ocr_det/params.py +++ b/deploy/hubserving/ocr_det/params.py @@ -27,7 +27,7 @@ def read_params(): # cfg.det_east_cover_thresh = 0.1 # cfg.det_east_nms_thresh = 0.2 - cfg.use_zero_copy_run = False cfg.use_pdserving = False + cfg.use_tensorrt = False return cfg diff --git a/deploy/hubserving/ocr_rec/params.py b/deploy/hubserving/ocr_rec/params.py index 6f428ecb..f8d29114 100644 --- a/deploy/hubserving/ocr_rec/params.py +++ b/deploy/hubserving/ocr_rec/params.py @@ -13,7 +13,7 @@ def read_params(): #params for text recognizer cfg.rec_algorithm = "CRNN" - cfg.rec_model_dir = "./inference/ch_ppocr_mobile_v1.1_rec_infer/" + cfg.rec_model_dir = "./inference/ch_ppocr_mobile_v2.0_rec_infer/" cfg.rec_image_shape = "3, 32, 320" cfg.rec_char_type = 'ch' @@ -23,7 +23,7 @@ def read_params(): cfg.rec_char_dict_path = "./ppocr/utils/ppocr_keys_v1.txt" cfg.use_space_char = True - cfg.use_zero_copy_run = False cfg.use_pdserving = False + cfg.use_tensorrt = False return cfg diff --git a/deploy/hubserving/ocr_system/params.py b/deploy/hubserving/ocr_system/params.py index a0e1960b..add46666 100755 --- a/deploy/hubserving/ocr_system/params.py +++ b/deploy/hubserving/ocr_system/params.py @@ -47,8 +47,8 @@ def read_params(): cfg.cls_batch_num = 30 cfg.cls_thresh = 0.9 - cfg.use_zero_copy_run = False cfg.use_pdserving = False + cfg.use_tensorrt = False cfg.drop_score = 0.5 return cfg From 554e57030b7c4a3fad397c159e185ee69c0863bb Mon Sep 17 00:00:00 2001 From: MissPenguin Date: Mon, 4 Jan 2021 12:47:32 +0800 Subject: [PATCH 53/60] Update quickstart.md --- doc/doc_ch/quickstart.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/doc/doc_ch/quickstart.md b/doc/doc_ch/quickstart.md index eabf1d91..3cc1b529 100644 --- a/doc/doc_ch/quickstart.md +++ b/doc/doc_ch/quickstart.md @@ -96,5 +96,5 @@ python3 tools/infer/predict_system.py --image_dir="./doc/imgs/11.jpg" --det_mode 此外，文档教程中也提供了中文OCR模型的其他预测部署方式： - [基于C++预测引擎推理](../../deploy/cpp_infer/readme.md) -- [服务部署](../../deploy/pdserving/readme.md) -- [端侧部署](../../deploy/lite/readme.md) +- [服务部署](../../deploy/hubserving/readme.md) +- [端侧部署（目前只支持静态图）](https://github.com/PaddlePaddle/PaddleOCR/tree/develop/deploy/lite) From c207f786bbbb867dff5af81c3717d00a7a4364d4 Mon Sep 17 00:00:00 2001 From: MissPenguin Date: Mon, 4 Jan 2021 12:48:28 +0800 Subject: [PATCH 54/60] Update quickstart.md --- doc/doc_ch/quickstart.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/doc/doc_ch/quickstart.md b/doc/doc_ch/quickstart.md index 3cc1b529..d9460989 100644 --- a/doc/doc_ch/quickstart.md +++ b/doc/doc_ch/quickstart.md @@ -96,5 +96,5 @@ python3 tools/infer/predict_system.py --image_dir="./doc/imgs/11.jpg" --det_mode 此外，文档教程中也提供了中文OCR模型的其他预测部署方式： - [基于C++预测引擎推理](../../deploy/cpp_infer/readme.md) -- [服务部署](../../deploy/hubserving/readme.md) +- [服务部署](../../deploy/hubserving) - [端侧部署（目前只支持静态图）](https://github.com/PaddlePaddle/PaddleOCR/tree/develop/deploy/lite) From a486350b8e73ab9b8aa86aba2a5c07649da9c2d2 Mon Sep 17 00:00:00 2001 From: MissPenguin Date: Mon, 4 Jan 2021 12:49:17 +0800 Subject: [PATCH 55/60] Update quickstart_en.md --- doc/doc_en/quickstart_en.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/doc/doc_en/quickstart_en.md b/doc/doc_en/quickstart_en.md index e351ecc6..a5c0881d 100644 --- a/doc/doc_en/quickstart_en.md +++ b/doc/doc_en/quickstart_en.md @@ -99,5 +99,5 @@ For more text detection and recognition tandem reasoning, please refer to the do In addition, the tutorial also provides other deployment methods for the Chinese OCR model: - [Server-side C++ inference](../../deploy/cpp_infer/readme_en.md) -- [Service deployment](../../deploy/pdserving/readme_en.md) -- [End-to-end deployment](../../deploy/lite/readme_en.md) +- [Service deployment](../../deploy/hubserving) +- [End-to-end deployment](https://github.com/PaddlePaddle/PaddleOCR/tree/develop/deploy/lite) From c91baf7cc66aa2db1911c99eafe68c6d99c2011a Mon Sep 17 00:00:00 2001 From: tink2123 Date: Mon, 4 Jan 2021 13:21:41 +0800 Subject: [PATCH 56/60] add faq, 210104 --- README_ch.md | 2 +- doc/doc_ch/FAQ.md | 72 +++++++++++++++++++++++++++++++++-------------- 2 files changed, 52 insertions(+), 22 deletions(-) diff --git a/README_ch.md b/README_ch.md index 3b0a2bfb..7aa43222 100755 --- a/README_ch.md +++ b/README_ch.md @@ -8,7 +8,7 @@ PaddleOCR同时支持动态图与静态图两种编程范式 - 静态图版本：develop分支 **近期更新** -- 2020.12.28 [FAQ](./doc/doc_ch/FAQ.md)新增5个高频问题，总数137个，每周一都会更新，欢迎大家持续关注。 +- 2021.1.4 [FAQ](./doc/doc_ch/FAQ.md)新增5个高频问题，总数142个，每周一都会更新，欢迎大家持续关注。 - 2020.12.15 更新数据合成工具[Style-Text](./StyleText/README_ch.md)，可以批量合成大量与目标场景类似的图像，在多个场景验证，效果明显提升。 - 2020.11.25 更新半自动标注工具[PPOCRLabel](./PPOCRLabel/README_ch.md)，辅助开发者高效完成标注任务，输出格式与PP-OCR训练任务完美衔接。 - 2020.9.22 更新PP-OCR技术文章，https://arxiv.org/abs/2009.09941 diff --git a/doc/doc_ch/FAQ.md b/doc/doc_ch/FAQ.md index 1abf30cc..2ad59f3d 100755 --- a/doc/doc_ch/FAQ.md +++ b/doc/doc_ch/FAQ.md @@ -9,45 +9,48 @@ ## PaddleOCR常见问题汇总(持续更新) -* [近期更新（2020.12.28）](#近期更新) +* [近期更新（2020.1.4）](#近期更新) * [【精选】OCR精选10个问题](#OCR精选10个问题) -* [【理论篇】OCR通用31个问题](#OCR通用问题) +* [【理论篇】OCR通用32个问题](#OCR通用问题) * [基础知识7题](#基础知识) * [数据集7题](#数据集2) - * [模型训练调优17题](#模型训练调优2) -* [【实战篇】PaddleOCR实战96个问题](#PaddleOCR实战问题) - * [使用咨询28题](#使用咨询) + * [模型训练调优18题](#模型训练调优2) +* [【实战篇】PaddleOCR实战100个问题](#PaddleOCR实战问题) + * [使用咨询31题](#使用咨询) * [数据集17题](#数据集3) * [模型训练调优26题](#模型训练调优3) - * [预测部署25题](#预测部署3) + * [预测部署26题](#预测部署3) -## 近期更新（2020.12.28） +## 近期更新（2021.1.4） -#### Q3.1.25: 使用dygraph分支，在docker中训练PaddleOCR的时候，数据路径没有任何问题，但是一直报错`reader rasied an exception`，这是为什么呢？ +#### Q2.3.18: 请问有哪些修改骨干网络的技巧？ -**A** 创建docker的时候，`/dev/shm`的默认大小为64M，如果使用多进程读取数据，共享内存可能不够，因此需要给`/dev/shm`分配更大的空间，在创建docker的时候，传入`--shm-size=8g`表示给`/dev/shm`分配8g的空间。 +**A**: 可以参考HS-ResNet这篇文章:https://arxiv.org/pdf/2010.07621.pdf -#### Q3.1.26: 在repo中没有找到Lite和PaddleServing相关的部署教程，这是在哪里呢？ +#### Q3.1.29: PPOCRLabel创建矩形框时只能拖出正方形，如何进行矩形标注？ -**A** 目前PaddleOCR的默认分支为dygraph，关于Lite和PaddleLite的动态图部署还在适配中，如果希望在Lite端或者使用PaddleServing部署，推荐使用develop分支（静态图）的代码。 +**A** 取消勾选：“编辑”-“正方形标注” -#### Q3.1.27: 如何可视化acc,loss曲线图,模型网络结构图等？ +#### Q3.1.30: Style-Text 如何不文字风格迁移，就像普通文本生成程序一样默认字体直接输出到分割的背景图？ -**A** 在配置文件里有`use_visualdl`的参数，设置为True即可，更多的使用命令可以参考：[VisualDL使用指南](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0-rc1/guides/03_VisualDL/visualdl.html)。 +**A** 使用image_synth模式会输出fake_bg.jpg，即为背景图。如果想要批量提取背景，可以稍微修改一下代码，将fake_bg保存下来即可。要修改的位置： +https://github.com/PaddlePaddle/PaddleOCR/blob/de3e2e7cd3b8b65ee02d7a41e570fa5b511a3c1d/StyleText/engine/synthesisers.py#L68 -#### Q3.1.28: 在使用StyleText数据合成工具的时候，报错`ModuleNotFoundError: No module named 'utils.config'`，这是为什么呢？ +#### Q3.1.31: 怎么输出网络结构以及每层的参数信息？ -**A** 有2个解决方案 -- 在StyleText路径下面设置PYTHONPATH：`export PYTHONPATH=./` -- 拉取最新的代码 - -#### Q3.3.26: PaddleOCR在训练的时候一直使用cosine_decay的学习率下降策略，这是为什么呢？ - -**A**: cosine_decay表示在训练的过程中，学习率按照cosine的变化趋势逐渐下降至0，在迭代轮数更长的情况下，比常量的学习率变化策略会有更好的收敛效果，因此在实际训练的时候，均采用了cosine_decay，来获得精度更高的模型。 +**A** 可以使用 `paddle.summary`， 具体参考:https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0-rc1/api/paddle/hapi/model_summary/summary_cn.html#summary。 +### Q3.4.26: 目前paddle hub serving 只支持 imgpath，如果我想用imgurl 去哪里改呢？ +**A**: 图片是在这里读取的：https://github.com/PaddlePaddle/PaddleOCR/blob/67ef25d593c4eabfaaceb22daade4577f53bed81/deploy/hubserving/ocr_system/module.py#L55， +可以参考下面的写法，将url path转化为np array（https://cloud.tencent.com/developer/article/1467840） +``` +response = request.urlopen('http://i1.whymtj.com/uploads/tu/201902/9999/52491ae4ba.jpg') +img_array = np.array(bytearray(response.read()), dtype=np.uint8) +img = cv.imdecode(img_array, -1) +``` ## 【精选】OCR精选10个问题 @@ -289,6 +292,10 @@ **A**：StyleText模型生成的数据主要用于OCR识别模型的训练。PaddleOCR目前识别模型的输入为32 x N，因此当前版本模型主要适用高度为32的数据。 建议要合成的数据尺寸设置为32 x N。尺寸相差不多的数据也可以生成，尺寸很大或很小的数据效果确实不佳。 +#### Q2.3.18: 请问有哪些修改骨干网络的技巧？ + +**A**: 可以参考HS-ResNet这篇文章:https://arxiv.org/pdf/2010.07621.pdf + ## 【实战篇】PaddleOCR实战问题 @@ -439,6 +446,19 @@ python3 -m pip install paddlepaddle-gpu==2.0.0rc1 -i https://mirror.baidu.com/py - 在StyleText路径下面设置PYTHONPATH：`export PYTHONPATH=./` - 拉取最新的代码 +#### Q3.1.29: PPOCRLabel创建矩形框时只能拖出正方形，如何进行矩形标注？ + +**A** 取消勾选：“编辑”-“正方形标注” + +#### Q3.1.30: Style-Text 如何不文字风格迁移，就像普通文本生成程序一样默认字体直接输出到分割的背景图？ + +**A** 使用image_synth模式会输出fake_bg.jpg，即为背景图。如果想要批量提取背景，可以稍微修改一下代码，将fake_bg保存下来即可。要修改的位置： +https://github.com/PaddlePaddle/PaddleOCR/blob/de3e2e7cd3b8b65ee02d7a41e570fa5b511a3c1d/StyleText/engine/synthesisers.py#L68 + +#### Q3.1.31: 怎么输出网络结构以及每层的参数信息？ + +**A** 可以使用 `paddle.summary`， 具体参考:https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0-rc1/api/paddle/hapi/model_summary/summary_cn.html#summary。 + ### 数据集 @@ -786,3 +806,13 @@ ps -axu | grep train.py | awk '{print $2}' | xargs kill -9 建议首先排查diff出现在检测模型还是识别模型，或者尝试换其他模型是否有类似的问题。 其次，检查python端和C++端数据处理部分是否存在差异，建议保存环境，更新PaddleOCR代码再试下。 如果更新代码或者更新代码都没能解决，建议在PaddleOCR微信群里或者issue中抛出您的问题。 + +### Q3.4.26: 目前paddle hub serving 只支持 imgpath，如果我想用imgurl 去哪里改呢？ + +**A**: 图片是在这里读取的：https://github.com/PaddlePaddle/PaddleOCR/blob/67ef25d593c4eabfaaceb22daade4577f53bed81/deploy/hubserving/ocr_system/module.py#L55， +可以参考下面的写法，将url path转化为np array（https://cloud.tencent.com/developer/article/1467840） +``` +response = request.urlopen('http://i1.whymtj.com/uploads/tu/201902/9999/52491ae4ba.jpg') +img_array = np.array(bytearray(response.read()), dtype=np.uint8) +img = cv.imdecode(img_array, -1) +``` From 10d7ecdbffefee6073790154052967a9dc2e4119 Mon Sep 17 00:00:00 2001 From: tink2123 Date: Mon, 4 Jan 2021 13:25:22 +0800 Subject: [PATCH 57/60] add faq, 210104 --- doc/doc_ch/FAQ.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/doc/doc_ch/FAQ.md b/doc/doc_ch/FAQ.md index 2ad59f3d..ca680388 100755 --- a/doc/doc_ch/FAQ.md +++ b/doc/doc_ch/FAQ.md @@ -9,7 +9,7 @@ ## PaddleOCR常见问题汇总(持续更新) -* [近期更新（2020.1.4）](#近期更新) +* [近期更新（2021.1.4）](#近期更新) * [【精选】OCR精选10个问题](#OCR精选10个问题) * [【理论篇】OCR通用32个问题](#OCR通用问题) * [基础知识7题](#基础知识) From 0ae781e19acf2f753a26cd763fb18cec8abb6408 Mon Sep 17 00:00:00 2001 From: tink2123 Date: Mon, 4 Jan 2021 15:28:49 +0800 Subject: [PATCH 58/60] add faq, 210104 --- README_ch.md | 4 ++-- doc/doc_ch/FAQ.md | 26 +++++++++++++++----------- 2 files changed, 17 insertions(+), 13 deletions(-) diff --git a/README_ch.md b/README_ch.md index 7aa43222..2a46ccea 100755 --- a/README_ch.md +++ b/README_ch.md @@ -101,8 +101,8 @@ PaddleOCR同时支持动态图与静态图两种编程范式 - [效果展示](#效果展示) - FAQ - [【精选】OCR精选10个问题](./doc/doc_ch/FAQ.md) - - [【理论篇】OCR通用31个问题](./doc/doc_ch/FAQ.md) - - [【实战篇】PaddleOCR实战96个问题](./doc/doc_ch/FAQ.md) + - [【理论篇】OCR通用32个问题](./doc/doc_ch/FAQ.md) + - [【实战篇】PaddleOCR实战101个问题](./doc/doc_ch/FAQ.md) - [技术交流群](#欢迎加入PaddleOCR技术交流群) - [参考文献](./doc/doc_ch/reference.md) - [许可证书](#许可证书) diff --git a/doc/doc_ch/FAQ.md b/doc/doc_ch/FAQ.md index ca680388..b1d78539 100755 --- a/doc/doc_ch/FAQ.md +++ b/doc/doc_ch/FAQ.md @@ -11,24 +11,20 @@ * [近期更新（2021.1.4）](#近期更新) * [【精选】OCR精选10个问题](#OCR精选10个问题) -* [【理论篇】OCR通用32个问题](#OCR通用问题) +* [【理论篇】OCR通用31个问题](#OCR通用问题) * [基础知识7题](#基础知识) * [数据集7题](#数据集2) - * [模型训练调优18题](#模型训练调优2) -* [【实战篇】PaddleOCR实战100个问题](#PaddleOCR实战问题) + * [模型训练调优17题](#模型训练调优2) +* [【实战篇】PaddleOCR实战101个问题](#PaddleOCR实战问题) * [使用咨询31题](#使用咨询) * [数据集17题](#数据集3) * [模型训练调优26题](#模型训练调优3) - * [预测部署26题](#预测部署3) + * [预测部署27题](#预测部署3) ## 近期更新（2021.1.4） -#### Q2.3.18: 请问有哪些修改骨干网络的技巧？ - -**A**: 可以参考HS-ResNet这篇文章:https://arxiv.org/pdf/2010.07621.pdf - #### Q3.1.29: PPOCRLabel创建矩形框时只能拖出正方形，如何进行矩形标注？ **A** 取消勾选：“编辑”-“正方形标注” @@ -52,6 +48,12 @@ img_array = np.array(bytearray(response.read()), dtype=np.uint8) img = cv.imdecode(img_array, -1) ``` +### Q3.4.27: C++ 端侧部署可以只对OCR的检测部署吗？ + +**A** 可以的，识别和检测模块是解耦的。如果想对检测部署，需要自己修改一下main函数， +只保留检测相关就可以:https://github.com/PaddlePaddle/PaddleOCR/blob/de3e2e7cd3b8b65ee02d7a41e570fa5b511a3c1d/deploy/cpp_infer/src/main.cpp#L72 + + ## 【精选】OCR精选10个问题 @@ -292,9 +294,6 @@ img = cv.imdecode(img_array, -1) **A**：StyleText模型生成的数据主要用于OCR识别模型的训练。PaddleOCR目前识别模型的输入为32 x N，因此当前版本模型主要适用高度为32的数据。 建议要合成的数据尺寸设置为32 x N。尺寸相差不多的数据也可以生成，尺寸很大或很小的数据效果确实不佳。 -#### Q2.3.18: 请问有哪些修改骨干网络的技巧？ - -**A**: 可以参考HS-ResNet这篇文章:https://arxiv.org/pdf/2010.07621.pdf @@ -816,3 +815,8 @@ response = request.urlopen('http://i1.whymtj.com/uploads/tu/201902/9999/52491ae4 img_array = np.array(bytearray(response.read()), dtype=np.uint8) img = cv.imdecode(img_array, -1) ``` + +### Q3.4.27: C++ 端侧部署可以只对OCR的检测部署吗？ + +**A** 可以的，识别和检测模块是解耦的。如果想对检测部署，需要自己修改一下main函数， +只保留检测相关就可以:https://github.com/PaddlePaddle/PaddleOCR/blob/de3e2e7cd3b8b65ee02d7a41e570fa5b511a3c1d/deploy/cpp_infer/src/main.cpp#L72 From 7ada917dfde361885229646ef93c0fc3bf46ec68 Mon Sep 17 00:00:00 2001 From: tink2123 Date: Mon, 4 Jan 2021 15:29:56 +0800 Subject: [PATCH 59/60] add faq, 210104 --- README_ch.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README_ch.md b/README_ch.md index 2a46ccea..163a19ed 100755 --- a/README_ch.md +++ b/README_ch.md @@ -101,7 +101,7 @@ PaddleOCR同时支持动态图与静态图两种编程范式 - [效果展示](#效果展示) - FAQ - [【精选】OCR精选10个问题](./doc/doc_ch/FAQ.md) - - [【理论篇】OCR通用32个问题](./doc/doc_ch/FAQ.md) + - [【理论篇】OCR通用31个问题](./doc/doc_ch/FAQ.md) - [【实战篇】PaddleOCR实战101个问题](./doc/doc_ch/FAQ.md) - [技术交流群](#欢迎加入PaddleOCR技术交流群) - [参考文献](./doc/doc_ch/reference.md) From c3e5522c38fb42ad721de2bae40425589f48b646 Mon Sep 17 00:00:00 2001 From: xiaoting <31891223+tink2123@users.noreply.github.com> Date: Tue, 5 Jan 2021 11:27:46 +0800 Subject: [PATCH 60/60] format faq --- doc/doc_ch/FAQ.md | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-) diff --git a/doc/doc_ch/FAQ.md b/doc/doc_ch/FAQ.md index b1d78539..da31d608 100755 --- a/doc/doc_ch/FAQ.md +++ b/doc/doc_ch/FAQ.md @@ -27,20 +27,20 @@ #### Q3.1.29: PPOCRLabel创建矩形框时只能拖出正方形，如何进行矩形标注？ -**A** 取消勾选：“编辑”-“正方形标注” +**A**： 取消勾选：“编辑”-“正方形标注” #### Q3.1.30: Style-Text 如何不文字风格迁移，就像普通文本生成程序一样默认字体直接输出到分割的背景图？ -**A** 使用image_synth模式会输出fake_bg.jpg，即为背景图。如果想要批量提取背景，可以稍微修改一下代码，将fake_bg保存下来即可。要修改的位置： +**A**： 使用image_synth模式会输出fake_bg.jpg，即为背景图。如果想要批量提取背景，可以稍微修改一下代码，将fake_bg保存下来即可。要修改的位置： https://github.com/PaddlePaddle/PaddleOCR/blob/de3e2e7cd3b8b65ee02d7a41e570fa5b511a3c1d/StyleText/engine/synthesisers.py#L68 #### Q3.1.31: 怎么输出网络结构以及每层的参数信息？ -**A** 可以使用 `paddle.summary`， 具体参考:https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0-rc1/api/paddle/hapi/model_summary/summary_cn.html#summary。 +**A**： 可以使用 `paddle.summary`， 具体参考:https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0-rc1/api/paddle/hapi/model_summary/summary_cn.html#summary。 -### Q3.4.26: 目前paddle hub serving 只支持 imgpath，如果我想用imgurl 去哪里改呢？ +#### Q3.4.26: 目前paddle hub serving 只支持 imgpath，如果我想用imgurl 去哪里改呢？ -**A**: 图片是在这里读取的：https://github.com/PaddlePaddle/PaddleOCR/blob/67ef25d593c4eabfaaceb22daade4577f53bed81/deploy/hubserving/ocr_system/module.py#L55， +**A**： 图片是在这里读取的：https://github.com/PaddlePaddle/PaddleOCR/blob/67ef25d593c4eabfaaceb22daade4577f53bed81/deploy/hubserving/ocr_system/module.py#L55， 可以参考下面的写法，将url path转化为np array（https://cloud.tencent.com/developer/article/1467840） ``` response = request.urlopen('http://i1.whymtj.com/uploads/tu/201902/9999/52491ae4ba.jpg') @@ -48,9 +48,9 @@ img_array = np.array(bytearray(response.read()), dtype=np.uint8) img = cv.imdecode(img_array, -1) ``` -### Q3.4.27: C++ 端侧部署可以只对OCR的检测部署吗？ +#### Q3.4.27: C++ 端侧部署可以只对OCR的检测部署吗？ -**A** 可以的，识别和检测模块是解耦的。如果想对检测部署，需要自己修改一下main函数， +**A**： 可以的，识别和检测模块是解耦的。如果想对检测部署，需要自己修改一下main函数， 只保留检测相关就可以:https://github.com/PaddlePaddle/PaddleOCR/blob/de3e2e7cd3b8b65ee02d7a41e570fa5b511a3c1d/deploy/cpp_infer/src/main.cpp#L72