# 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. 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__, '../..'))) sys.path.append(os.path.abspath(os.path.join(__dir__, '../../..'))) os.environ["FLAGS_allocator_strategy"] = 'auto_growth' # print("sys.path", sys.path) import cv2 import copy import numpy as np import time from PIL import Image os.environ['FLAGS_eager_delete_tensor_gb'] = '0' import utility as utility # import ocr.tools.infer.predict_rec as predict_rec import ocr.tools.infer.predict_rec_pytorch as predict_rec # pytorch rec model # import ocr.tools.infer.predict_det as predict_det import ocr.tools.infer.predict_det_pytorch as predict_det # pytorch det model import ocr.tools.infer.predict_cls as predict_cls from ocr.ppocr.utils.utility import get_image_file_list, check_and_read_gif from ocr.ppocr.utils.logging import get_logger from ocr.tools.infer.utility import draw_ocr_box_txt from format_convert.utils import has_intersection, log from format_convert import _global logger = get_logger() class TextSystem(object): def __init__(self, args): self.text_detector = predict_det.TextDetector(args) self.text_recognizer = predict_rec.TextRecognizer(args) self.use_angle_cls = args.use_angle_cls self.drop_score = args.drop_score if self.use_angle_cls: self.text_classifier = predict_cls.TextClassifier(args) def get_rotate_crop_image(self, img, points): ''' img_height, img_width = img.shape[0:2] left = int(np.min(points[:, 0])) right = int(np.max(points[:, 0])) top = int(np.min(points[:, 1])) bottom = int(np.max(points[:, 1])) img_crop = img[top:bottom, left:right, :].copy() points[:, 0] = points[:, 0] - left points[:, 1] = points[:, 1] - top ''' # img_crop_width = int( # max( # np.linalg.norm(points[0] - points[1]), # np.linalg.norm(points[2] - points[3]))) # img_crop_height = int( # max( # np.linalg.norm(points[0] - points[3]), # np.linalg.norm(points[1] - points[2]))) # pts_std = np.float32([[0, 0], [img_crop_width, 0], # [img_crop_width, img_crop_height], # [0, img_crop_height]]) # M = cv2.getPerspectiveTransform(points, pts_std) # dst_img = cv2.warpPerspective( # img, # M, (img_crop_width, img_crop_height), # borderMode=cv2.BORDER_REPLICATE, # flags=cv2.INTER_CUBIC) # print('dst_img.shape', dst_img.shape) # # print('points', points) w = abs(points[2][0] - points[0][0]) h = abs(points[2][1] - points[0][1]) dst_img = img[int(points[0][1]):int(points[0][1] + h), int(points[0][0]):int(points[0][0] + w), :] # print('dst_img.shape2', dst_img.shape) # cv2.imshow('dst_img', dst_img) # cv2.waitKey(0) # dst_img_height, dst_img_width = dst_img.shape[0:2] # # if dst_img_height * 1.0 / dst_img_width >= 1.5: # if dst_img_height * 1.0 / dst_img_width >= 2.0: # dst_img = np.rot90(dst_img) return dst_img def print_draw_crop_rec_res(self, img_crop_list, rec_res): bbox_num = len(img_crop_list) for bno in range(bbox_num): cv2.imwrite("./output/img_crop_%d.jpg" % bno, img_crop_list[bno]) logger.info(bno, rec_res[bno]) def __call__(self, img): # print('into TextSystem __call__') # cv2.imshow('img',img) # cv2.waitKey(0) ori_im = img.copy() dt_boxes, elapse = self.text_detector(img) logger.info("dt_boxes num : {}, elapse : {}".format( len(dt_boxes), elapse)) if dt_boxes is None: return [], [] temp_list = [] # print('dt_boxes', type(dt_boxes)) # print('dt_boxes.shape', dt_boxes.shape) # 过滤一些比例离谱的box for b in dt_boxes: w = b[2][0] - b[0][0] h = b[2][1] - b[0][1] if h == 0 or w == 0 \ or h >= 10000 or w >= 10000 \ or w / h <= 0.5 or w / h >= 100: continue temp_list.append(b) if not temp_list: return [], [] dt_boxes = np.array(temp_list) # print('dt_boxes.shape2', dt_boxes.shape) # show # for b in dt_boxes: # p1 = [int(x) for x in b[0]] # p2 = [int(x) for x in b[2]] # cv2.rectangle(img, p1, p2, (0, 0, 255)) # cv2.namedWindow('img', cv2.WINDOW_NORMAL) # cv2.imshow('img', img) # cv2.waitKey(0) # # 检测过多单字box，返回None # if len(dt_boxes) >= 150: # short_box_cnt = 0 # long_box_cnt = 0 # for b in dt_boxes: # w = b[2][0] - b[0][0] # h = b[2][1] - b[0][1] # if w / h < 1.3: # short_box_cnt += 1 # if w / h >= 3: # long_box_cnt += 1 # print('dt_boxes', w, h, round(w/h, 3)) # # print('short_box_cnt, len(dt_boxes)', short_box_cnt, len(dt_boxes)) # log('short_box_cnt, long_box_cnt, len(dt_boxes) ' + str([short_box_cnt, long_box_cnt, len(dt_boxes)])) # if short_box_cnt >= 2/3 * len(dt_boxes) and long_box_cnt < 10: # # print('short_box_cnt >= 2/3 * len(dt_boxes), return None') # log('short_box_cnt >= 2/3 * len(dt_boxes), return None. ' + str([short_box_cnt, long_box_cnt, len(dt_boxes)])) # return [], [] img_crop_list = [] dt_boxes = sorted_boxes(dt_boxes) for bno in range(len(dt_boxes)): tmp_box = copy.deepcopy(dt_boxes[bno]) img_crop = self.get_rotate_crop_image(ori_im, tmp_box) img_crop_list.append(img_crop) # print('system len(img_crop_list)', len(img_crop_list)) # for img in img_crop_list: # if img.shape[1] / img.shape[0] <= 0.5: # print('system img.shape[1] / img.shape[0] <= 0.5', img.shape) if self.use_angle_cls: img_crop_list, angle_list, elapse = self.text_classifier( img_crop_list) logger.info("cls num : {}, elapse : {}".format( len(img_crop_list), elapse)) rec_res, elapse = self.text_recognizer(img_crop_list) logger.info("rec_res num : {}, elapse : {}".format( len(rec_res), elapse)) # self.print_draw_crop_rec_res(img_crop_list, rec_res) filter_boxes, filter_rec_res = [], [] # dt_boxes 上下重合检测框修正 # t1 = time.time() dt_boxes = boxex_points_fixup(dt_boxes) # print("boxex_points_fixup cost:",time.time()-t1) for box, rec_reuslt in zip(dt_boxes, rec_res): text, score = rec_reuslt if score >= self.drop_score: filter_boxes.append(box) filter_rec_res.append(rec_reuslt) return filter_boxes, filter_rec_res def boxex_points_fixup(dt_boxes): # 检查框全部转换为矩形 # for i in range(len(dt_boxes)): # box1 = dt_boxes[i] # x_list = [box1[0][0],box1[1][0],box1[2][0],box1[3][0]] # y_list = [box1[0][1],box1[1][1],box1[2][1],box1[3][1]] # x_max = max(x_list) # x_min = min(x_list) # y_max = max(y_list) # y_min = min(y_list) # dt_boxes[i] = np.array([[x_min,y_min],[x_max,y_min],[x_max,y_max],[x_min,y_max]]) for i in range(len(dt_boxes)): box1 = dt_boxes[i] box1_point3 = box1[2] box1_point4 = box1[3] # 四边形底边的两点坐标 bottom_line = (min(box1_point3[0], box1_point4[0]), max(box1_point3[0], box1_point4[0])) bottom_line_len = abs(bottom_line[1] - bottom_line[0]) for j in range(i + 1, len(dt_boxes)): box2 = dt_boxes[j] box2_point1 = box2[0] box2_point2 = box2[1] # 四边形顶边的两点坐标 top_line = (min(box2_point1[0], box2_point2[0]), max(box2_point1[0], box2_point2[0])) top_line_len = abs(top_line[1] - top_line[0]) if has_intersection(box1, box2): # 四边形框是否有交集 if not (min(top_line) >= max(bottom_line) or min(bottom_line) >= max(top_line)): # x轴方向上有交集 # 求重合部分y中间值 mid_y = ((box2_point1[1] + box2_point2[1]) / 2 + (box1_point3[1] + box1_point4[1]) / 2) // 2 if not mid_y: continue max_line_len = max(bottom_line_len, top_line_len) cross_line_len = bottom_line_len + top_line_len - \ (max(bottom_line[1], bottom_line[0], top_line[1], top_line[0]) - min( bottom_line[1], bottom_line[0], top_line[1], top_line[0])) # print(cross_line_len,max_line_len,cross_line_len/max_line_len) if cross_line_len / max_line_len >= 0.55: # 重合比例 box1[2] = [box1_point3[0], mid_y] box1[3] = [box1_point4[0], mid_y] box2[0] = [box2_point1[0], mid_y] box2[1] = [box2_point2[0], mid_y] break return dt_boxes def sorted_boxes(dt_boxes): """ Sort text boxes in order from top to bottom, left to right args: dt_boxes(array):detected text boxes with shape [4, 2] return: sorted boxes(array) with shape [4, 2] """ num_boxes = dt_boxes.shape[0] sorted_boxes = sorted(dt_boxes, key=lambda x: (x[0][1], x[0][0])) _boxes = list(sorted_boxes) for i in range(num_boxes - 1): if abs(_boxes[i + 1][0][1] - _boxes[i][0][1]) < 10 and \ (_boxes[i + 1][0][0] < _boxes[i][0][0]): tmp = _boxes[i] _boxes[i] = _boxes[i + 1] _boxes[i + 1] = tmp return _boxes def main(args): image_file_list = get_image_file_list(args.image_dir) text_sys = TextSystem(args) is_visualize = True font_path = args.vis_font_path drop_score = args.drop_score for image_file in image_file_list: img, flag = check_and_read_gif(image_file) if not flag: img = cv2.imread(image_file) if img is None: logger.info("error in loading image:{}".format(image_file)) continue starttime = time.time() dt_boxes, rec_res = text_sys(img) elapse = time.time() - starttime logger.info("Predict time of %s: %.3fs" % (image_file, elapse)) for text, score in rec_res: logger.info("{}, {:.3f}".format(text, score)) if is_visualize: image = Image.fromarray(cv2.cvtColor(img, cv2.COLOR_BGR2RGB)) boxes = dt_boxes txts = [rec_res[i][0] for i in range(len(rec_res))] scores = [rec_res[i][1] for i in range(len(rec_res))] draw_img = draw_ocr_box_txt( image, boxes, txts, scores, drop_score=drop_score, font_path=font_path) draw_img_save = "./inference_results/" if not os.path.exists(draw_img_save): os.makedirs(draw_img_save) cv2.imwrite( os.path.join(draw_img_save, os.path.basename(image_file)), draw_img[:, :, ::-1]) logger.info("The visualized image saved in {}".format( os.path.join(draw_img_save, os.path.basename(image_file)))) if __name__ == "__main__": main(utility.parse_args()) pass