PaddleOCR/tools/infer/predict_rec.py

192 lines
7.8 KiB
Python
Raw Normal View History

2020-05-10 16:26:57 +08:00
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# 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.
2020-06-12 13:49:24 +08:00
import os
import sys
__dir__ = os.path.dirname(os.path.abspath(__file__))
2020-06-12 13:49:24 +08:00
sys.path.append(__dir__)
sys.path.append(os.path.abspath(os.path.join(__dir__, '../..')))
2020-05-10 16:26:57 +08:00
import cv2
import copy
import numpy as np
import math
import time
import paddle.fluid as fluid
import tools.infer.utility as utility
from ppocr.utils.utility import initial_logger
logger = initial_logger()
from ppocr.utils.utility import get_image_file_list, check_and_read_gif
2020-05-10 16:26:57 +08:00
from ppocr.utils.character import CharacterOps
class TextRecognizer(object):
def __init__(self, args):
self.predictor, self.input_tensor, self.output_tensors =\
utility.create_predictor(args, mode="rec")
self.rec_image_shape = [int(v) for v in args.rec_image_shape.split(",")]
2020-05-15 22:07:18 +08:00
self.character_type = args.rec_char_type
self.rec_batch_num = args.rec_batch_num
2020-06-03 17:38:44 +08:00
self.rec_algorithm = args.rec_algorithm
2020-08-24 10:11:17 +08:00
self.use_zero_copy_run = args.use_zero_copy_run
2020-06-30 11:18:49 +08:00
char_ops_params = {
"character_type": args.rec_char_type,
2020-07-07 14:13:13 +08:00
"character_dict_path": args.rec_char_dict_path,
2020-08-16 12:53:26 +08:00
"use_space_char": args.use_space_char,
"max_text_length": args.max_text_length
2020-06-30 11:18:49 +08:00
}
2020-06-03 17:38:44 +08:00
if self.rec_algorithm != "RARE":
char_ops_params['loss_type'] = 'ctc'
2020-06-04 19:41:42 +08:00
self.loss_type = 'ctc'
2020-06-03 17:38:44 +08:00
else:
char_ops_params['loss_type'] = 'attention'
2020-06-04 19:41:42 +08:00
self.loss_type = 'attention'
2020-05-10 16:26:57 +08:00
self.char_ops = CharacterOps(char_ops_params)
def resize_norm_img(self, img, max_wh_ratio):
2020-05-10 16:26:57 +08:00
imgC, imgH, imgW = self.rec_image_shape
assert imgC == img.shape[2]
if self.character_type == "ch":
2020-06-30 14:47:24 +08:00
imgW = int((32 * max_wh_ratio))
h, w = img.shape[:2]
ratio = w / float(h)
if math.ceil(imgH * ratio) > imgW:
resized_w = imgW
else:
resized_w = int(math.ceil(imgH * ratio))
2020-06-30 11:18:49 +08:00
resized_image = cv2.resize(img, (resized_w, imgH))
2020-05-10 16:26:57 +08:00
resized_image = resized_image.astype('float32')
resized_image = resized_image.transpose((2, 0, 1)) / 255
resized_image -= 0.5
resized_image /= 0.5
padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
padding_im[:, :, 0:resized_w] = resized_image
return padding_im
def __call__(self, img_list):
img_num = len(img_list)
# Calculate the aspect ratio of all text bars
width_list = []
for img in img_list:
width_list.append(img.shape[1] / float(img.shape[0]))
2020-06-27 23:29:29 +08:00
# Sorting can speed up the recognition process
indices = np.argsort(np.array(width_list))
# rec_res = []
rec_res = [['', 0.0]] * img_num
batch_num = self.rec_batch_num
2020-05-10 16:26:57 +08:00
predict_time = 0
for beg_img_no in range(0, img_num, batch_num):
end_img_no = min(img_num, beg_img_no + batch_num)
norm_img_batch = []
max_wh_ratio = 0
2020-05-10 16:26:57 +08:00
for ino in range(beg_img_no, end_img_no):
# h, w = img_list[ino].shape[0:2]
h, w = img_list[indices[ino]].shape[0:2]
wh_ratio = w * 1.0 / h
max_wh_ratio = max(max_wh_ratio, wh_ratio)
for ino in range(beg_img_no, end_img_no):
# norm_img = self.resize_norm_img(img_list[ino], max_wh_ratio)
2020-06-30 11:18:49 +08:00
norm_img = self.resize_norm_img(img_list[indices[ino]],
max_wh_ratio)
2020-05-10 16:26:57 +08:00
norm_img = norm_img[np.newaxis, :]
norm_img_batch.append(norm_img)
norm_img_batch = np.concatenate(norm_img_batch)
norm_img_batch = norm_img_batch.copy()
starttime = time.time()
2020-08-24 10:11:17 +08:00
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])
2020-06-03 13:44:07 +08:00
2020-06-04 19:41:42 +08:00
if self.loss_type == "ctc":
2020-06-03 13:44:07 +08:00
rec_idx_batch = self.output_tensors[0].copy_to_cpu()
rec_idx_lod = self.output_tensors[0].lod()[0]
predict_batch = self.output_tensors[1].copy_to_cpu()
predict_lod = self.output_tensors[1].lod()[0]
elapse = time.time() - starttime
predict_time += elapse
for rno in range(len(rec_idx_lod) - 1):
beg = rec_idx_lod[rno]
end = rec_idx_lod[rno + 1]
rec_idx_tmp = rec_idx_batch[beg:end, 0]
preds_text = self.char_ops.decode(rec_idx_tmp)
beg = predict_lod[rno]
end = predict_lod[rno + 1]
probs = predict_batch[beg:end, :]
ind = np.argmax(probs, axis=1)
blank = probs.shape[1]
valid_ind = np.where(ind != (blank - 1))[0]
2020-07-01 13:30:03 +08:00
if len(valid_ind) == 0:
continue
2020-08-03 22:22:08 +08:00
score = np.mean(probs[valid_ind, ind[valid_ind]])
# rec_res.append([preds_text, score])
rec_res[indices[beg_img_no + rno]] = [preds_text, score]
2020-06-03 13:44:07 +08:00
else:
rec_idx_batch = self.output_tensors[0].copy_to_cpu()
predict_batch = self.output_tensors[1].copy_to_cpu()
2020-06-03 17:38:44 +08:00
elapse = time.time() - starttime
predict_time += elapse
2020-06-03 13:44:07 +08:00
for rno in range(len(rec_idx_batch)):
end_pos = np.where(rec_idx_batch[rno, :] == 1)[0]
if len(end_pos) <= 1:
preds = rec_idx_batch[rno, 1:]
score = np.mean(predict_batch[rno, 1:])
else:
preds = rec_idx_batch[rno, 1:end_pos[1]]
score = np.mean(predict_batch[rno, 1:end_pos[1]])
preds_text = self.char_ops.decode(preds)
# rec_res.append([preds_text, score])
rec_res[indices[beg_img_no + rno]] = [preds_text, score]
2020-06-03 13:44:07 +08:00
2020-05-10 16:26:57 +08:00
return rec_res, predict_time
def main(args):
2020-05-15 22:07:18 +08:00
image_file_list = get_image_file_list(args.image_dir)
2020-05-10 16:26:57 +08:00
text_recognizer = TextRecognizer(args)
valid_image_file_list = []
img_list = []
for image_file in image_file_list:
2020-07-28 11:18:48 +08:00
img, flag = check_and_read_gif(image_file)
if not flag:
img = cv2.imread(image_file)
2020-05-10 16:26:57 +08:00
if img is None:
logger.info("error in loading image:{}".format(image_file))
continue
valid_image_file_list.append(image_file)
img_list.append(img)
2020-06-03 13:44:07 +08:00
try:
rec_res, predict_time = text_recognizer(img_list)
2020-06-04 19:41:42 +08:00
except Exception as e:
print(e)
2020-06-03 13:44:07 +08:00
logger.info(
2020-06-04 19:41:42 +08:00
"ERROR!!!! \n"
"Please read the FAQhttps://github.com/PaddlePaddle/PaddleOCR#faq \n"
"If your model has tps module: "
"TPS does not support variable shape.\n"
2020-06-05 11:29:02 +08:00
"Please set --rec_image_shape='3,32,100' and --rec_char_type='en' ")
2020-06-03 13:44:07 +08:00
exit()
2020-05-10 16:26:57 +08:00
for ino in range(len(img_list)):
print("Predicts of %s:%s" % (valid_image_file_list[ino], rec_res[ino]))
print("Total predict time for %d images:%.3f" %
(len(img_list), predict_time))
if __name__ == "__main__":
main(utility.parse_args())