From 7a054c854b8253a2a088c107e55de20a3f207a26 Mon Sep 17 00:00:00 2001 From: LDOUBLEV Date: Mon, 1 Feb 2021 06:27:56 +0000 Subject: [PATCH] rare doc and opt post_process --- doc/doc_ch/algorithm_overview.md | 5 +++- doc/doc_ch/recognition.md | 2 ++ doc/doc_en/algorithm_overview_en.md | 4 ++- doc/doc_en/recognition_en.md | 3 +++ ppocr/postprocess/rec_postprocess.py | 37 +++++++++++++++++++++++++--- 5 files changed, 45 insertions(+), 6 deletions(-) diff --git a/doc/doc_ch/algorithm_overview.md b/doc/doc_ch/algorithm_overview.md index abbc5da4..4ff7482c 100755 --- a/doc/doc_ch/algorithm_overview.md +++ b/doc/doc_ch/algorithm_overview.md @@ -40,7 +40,7 @@ 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))[7]（ppocr推荐） - [x] Rosetta([paper](https://arxiv.org/abs/1910.05085))[10] - [x] STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html))[11] -- [ ] RARE([paper](https://arxiv.org/abs/1603.03915v1))[12] coming soon +- [x] RARE([paper](https://arxiv.org/abs/1603.03915v1))[12] - [x] SRN([paper](https://arxiv.org/abs/2003.12294))[5] 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： @@ -53,6 +53,9 @@ PaddleOCR基于动态图开源的文本识别算法列表： |CRNN|MobileNetV3|79.97%|rec_mv3_none_bilstm_ctc|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_none_bilstm_ctc_v2.0_train.tar)| |StarNet|Resnet34_vd|84.44%|rec_r34_vd_tps_bilstm_ctc|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_tps_bilstm_ctc_v2.0_train.tar)| |StarNet|MobileNetV3|81.42%|rec_mv3_tps_bilstm_ctc|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_tps_bilstm_ctc_v2.0_train.tar)| +|RARE|MobileNetV3|82.5|rec_mv3_tps_bilstm_att||[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_tps_bilstm_att_v2.0_train.tar)| +|RARE|Resnet34_vd|83.6|rec_r34_vd_tps_bilstm_att||[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_tps_bilstm_att_v2.0_train.tar)| |SRN|Resnet50_vd_fpn| 88.52% | rec_r50fpn_vd_none_srn | [下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r50_vd_srn_train.tar) | + PaddleOCR文本识别算法的训练和使用请参考文档教程中[模型训练/评估中的文本识别部分](./recognition.md)。 diff --git a/doc/doc_ch/recognition.md b/doc/doc_ch/recognition.md index bc877ab7..f36e8019 100644 --- a/doc/doc_ch/recognition.md +++ b/doc/doc_ch/recognition.md @@ -201,6 +201,8 @@ PaddleOCR支持训练和评估交替进行, 可以在 `configs/rec/rec_icdar15_t | rec_mv3_none_none_ctc.yml | Rosetta | Mobilenet_v3 large 0.5 | None | None | ctc | | rec_r34_vd_none_bilstm_ctc.yml | CRNN | Resnet34_vd | None | BiLSTM | ctc | | rec_r34_vd_none_none_ctc.yml | Rosetta | Resnet34_vd | None | None | ctc | +| rec_mv3_tps_bilstm_att.yml | CRNN | Mobilenet_v3 | TPS | BiLSTM | att | +| rec_r34_vd_tps_bilstm_att.yml | CRNN | Resnet34_vd | TPS | BiLSTM | att | | rec_r50fpn_vd_none_srn.yml | SRN | Resnet50_fpn_vd | None | rnn | srn | 训练中文数据，推荐使用[rec_chinese_lite_train_v2.0.yml](../../configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml)，如您希望尝试其他算法在中文数据集上的效果，请参考下列说明修改配置文件： diff --git a/doc/doc_en/algorithm_overview_en.md b/doc/doc_en/algorithm_overview_en.md index 7d7896e7..423fe807 100755 --- a/doc/doc_en/algorithm_overview_en.md +++ b/doc/doc_en/algorithm_overview_en.md @@ -42,7 +42,7 @@ PaddleOCR open-source text recognition algorithms list: - [x] CRNN([paper](https://arxiv.org/abs/1507.05717))[7] - [x] Rosetta([paper](https://arxiv.org/abs/1910.05085))[10] - [x] STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html))[11] -- [ ] RARE([paper](https://arxiv.org/abs/1603.03915v1))[12] coming soon +- [x] RARE([paper](https://arxiv.org/abs/1603.03915v1))[12] - [x] SRN([paper](https://arxiv.org/abs/2003.12294))[5] 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: @@ -55,6 +55,8 @@ Refer to [DTRB](https://arxiv.org/abs/1904.01906), the training and evaluation r |CRNN|MobileNetV3|79.97%|rec_mv3_none_bilstm_ctc|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_none_bilstm_ctc_v2.0_train.tar)| |StarNet|Resnet34_vd|84.44%|rec_r34_vd_tps_bilstm_ctc|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_tps_bilstm_ctc_v2.0_train.tar)| |StarNet|MobileNetV3|81.42%|rec_mv3_tps_bilstm_ctc|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_tps_bilstm_ctc_v2.0_train.tar)| +|RARE|MobileNetV3|82.5|rec_mv3_tps_bilstm_att||[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_tps_bilstm_att_v2.0_train.tar)| +|RARE|Resnet34_vd|83.6|rec_r34_vd_tps_bilstm_att||[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_tps_bilstm_att_v2.0_train.tar)| |SRN|Resnet50_vd_fpn| 88.52% | rec_r50fpn_vd_none_srn |[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r50_vd_srn_train.tar)| Please refer to the document for training guide and use of PaddleOCR text recognition algorithms [Text recognition model training/evaluation/prediction](./recognition_en.md) diff --git a/doc/doc_en/recognition_en.md b/doc/doc_en/recognition_en.md index f29703d1..c2ff2022 100644 --- a/doc/doc_en/recognition_en.md +++ b/doc/doc_en/recognition_en.md @@ -195,8 +195,11 @@ If the evaluation set is large, the test will be time-consuming. It is recommend | rec_mv3_none_none_ctc.yml | Rosetta | Mobilenet_v3 large 0.5 | None | None | ctc | | rec_r34_vd_none_bilstm_ctc.yml | CRNN | Resnet34_vd | None | BiLSTM | ctc | | rec_r34_vd_none_none_ctc.yml | Rosetta | Resnet34_vd | None | None | ctc | +| rec_mv3_tps_bilstm_att.yml | CRNN | Mobilenet_v3 | TPS | BiLSTM | att | +| rec_r34_vd_tps_bilstm_att.yml | CRNN | Resnet34_vd | TPS | BiLSTM | att | | rec_r50fpn_vd_none_srn.yml | SRN | Resnet50_fpn_vd | None | rnn | srn | + For training Chinese data, it is recommended to use [rec_chinese_lite_train_v2.0.yml](../../configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml). If you want to try the result of other algorithms on the Chinese data set, please refer to the following instructions to modify the configuration file: co diff --git a/ppocr/postprocess/rec_postprocess.py b/ppocr/postprocess/rec_postprocess.py index 2b82750f..d7e658f3 100644 --- a/ppocr/postprocess/rec_postprocess.py +++ b/ppocr/postprocess/rec_postprocess.py @@ -143,6 +143,35 @@ class AttnLabelDecode(BaseRecLabelDecode): dict_character = [self.beg_str] + dict_character + [self.end_str] return dict_character + def decode(self, text_index, text_prob=None, is_remove_duplicate=False): + """ convert text-index into text-label. """ + result_list = [] + ignored_tokens = self.get_ignored_tokens() + [beg_idx, end_idx] = self.get_ignored_tokens() + batch_size = len(text_index) + for batch_idx in range(batch_size): + char_list = [] + conf_list = [] + for idx in range(len(text_index[batch_idx])): + if text_index[batch_idx][idx] in ignored_tokens: + continue + if int(text_index[batch_idx][idx]) == int(end_idx): + break + if is_remove_duplicate: + # only for predict + if idx > 0 and text_index[batch_idx][idx - 1] == text_index[ + batch_idx][idx]: + continue + char_list.append(self.character[int(text_index[batch_idx][ + idx])]) + if text_prob is not None: + conf_list.append(text_prob[batch_idx][idx]) + else: + conf_list.append(1) + text = ''.join(char_list) + result_list.append((text, np.mean(conf_list))) + return result_list + def __call__(self, preds, label=None, *args, **kwargs): """ text = self.decode(text) @@ -157,10 +186,10 @@ class AttnLabelDecode(BaseRecLabelDecode): preds_idx = preds.argmax(axis=2) preds_prob = preds.max(axis=2) - text = self.decode(preds_idx, preds_prob, is_remove_duplicate=True) + text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False) if label is None: return text - label = self.decode(label, is_remove_duplicate=True) + label = self.decode(label, is_remove_duplicate=False) return text, label def encoder(self, labels, labels_length): @@ -226,12 +255,12 @@ class SRNLabelDecode(BaseRecLabelDecode): text = self.decode(preds_idx, preds_prob) if label is None: - text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False) + text = self.decode(preds_idx, preds_prob, is_remove_duplicate=True) return text label = self.decode(label) return text, label - def decode(self, text_index, text_prob=None, is_remove_duplicate=False): + def decode(self, text_index, text_prob=None, is_remove_duplicate=True): """ convert text-index into text-label. """ result_list = [] ignored_tokens = self.get_ignored_tokens()