predict_system.py 6.8 KB
Newer Older
L
LDOUBLEV 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13
# 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.
14 15
import os
import sys
W
WenmuZhou 已提交
16

17
__dir__ = os.path.dirname(os.path.abspath(__file__))
18
sys.path.append(__dir__)
19
sys.path.append(os.path.abspath(os.path.join(__dir__, '../..')))
L
LDOUBLEV 已提交
20

L
LDOUBLEV 已提交
21 22
os.environ["FLAGS_allocator_strategy"] = 'auto_growth'

L
LDOUBLEV 已提交
23 24 25 26
import cv2
import copy
import numpy as np
import time
L
LDOUBLEV 已提交
27
from PIL import Image
W
WenmuZhou 已提交
28 29 30
import tools.infer.utility as utility
import tools.infer.predict_rec as predict_rec
import tools.infer.predict_det as predict_det
W
WenmuZhou 已提交
31
import tools.infer.predict_cls as predict_cls
W
WenmuZhou 已提交
32 33
from ppocr.utils.utility import get_image_file_list, check_and_read_gif
from ppocr.utils.logging import get_logger
W
WenmuZhou 已提交
34
from tools.infer.utility import draw_ocr_box_txt
L
LDOUBLEV 已提交
35

W
WenmuZhou 已提交
36 37
logger = get_logger()

L
LDOUBLEV 已提交
38 39 40 41 42

class TextSystem(object):
    def __init__(self, args):
        self.text_detector = predict_det.TextDetector(args)
        self.text_recognizer = predict_rec.TextRecognizer(args)
W
WenmuZhou 已提交
43
        self.use_angle_cls = args.use_angle_cls
W
WenmuZhou 已提交
44
        self.drop_score = args.drop_score
W
WenmuZhou 已提交
45 46
        if self.use_angle_cls:
            self.text_classifier = predict_cls.TextClassifier(args)
L
LDOUBLEV 已提交
47 48

    def get_rotate_crop_image(self, img, points):
49
        '''
L
LDOUBLEV 已提交
50 51 52 53 54 55 56 57
        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
58
        '''
L
LDOUBLEV 已提交
59 60 61 62 63 64 65 66 67
        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],
68 69
                              [img_crop_width, img_crop_height],
                              [0, img_crop_height]])
L
LDOUBLEV 已提交
70
        M = cv2.getPerspectiveTransform(points, pts_std)
L
LDOUBLEV 已提交
71 72 73 74 75
        dst_img = cv2.warpPerspective(
            img,
            M, (img_crop_width, img_crop_height),
            borderMode=cv2.BORDER_REPLICATE,
            flags=cv2.INTER_CUBIC)
L
LDOUBLEV 已提交
76 77 78 79 80 81 82 83 84
        dst_img_height, dst_img_width = dst_img.shape[0:2]
        if dst_img_height * 1.0 / dst_img_width >= 1.5:
            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])
W
WenmuZhou 已提交
85
            logger.info(bno, rec_res[bno])
L
LDOUBLEV 已提交
86 87 88 89

    def __call__(self, img):
        ori_im = img.copy()
        dt_boxes, elapse = self.text_detector(img)
W
WenmuZhou 已提交
90 91
        logger.info("dt_boxes num : {}, elapse : {}".format(
            len(dt_boxes), elapse))
L
LDOUBLEV 已提交
92 93 94
        if dt_boxes is None:
            return None, None
        img_crop_list = []
95 96 97

        dt_boxes = sorted_boxes(dt_boxes)

L
LDOUBLEV 已提交
98 99 100 101
        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)
W
WenmuZhou 已提交
102 103 104
        if self.use_angle_cls:
            img_crop_list, angle_list, elapse = self.text_classifier(
                img_crop_list)
W
WenmuZhou 已提交
105
            logger.info("cls num  : {}, elapse : {}".format(
W
WenmuZhou 已提交
106 107
                len(img_crop_list), elapse))

L
LDOUBLEV 已提交
108
        rec_res, elapse = self.text_recognizer(img_crop_list)
W
WenmuZhou 已提交
109 110
        logger.info("rec_res num  : {}, elapse : {}".format(
            len(rec_res), elapse))
111
        # self.print_draw_crop_rec_res(img_crop_list, rec_res)
W
WenmuZhou 已提交
112 113 114 115 116 117 118
        filter_boxes, filter_rec_res = [], []
        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
L
LDOUBLEV 已提交
119 120


121 122 123 124
def sorted_boxes(dt_boxes):
    """
    Sort text boxes in order from top to bottom, left to right
    args:
T
tink2123 已提交
125
        dt_boxes(array):detected text boxes with shape [4, 2]
126 127 128 129
    return:
        sorted boxes(array) with shape [4, 2]
    """
    num_boxes = dt_boxes.shape[0]
130
    sorted_boxes = sorted(dt_boxes, key=lambda x: (x[0][1], x[0][0]))
131 132 133
    _boxes = list(sorted_boxes)

    for i in range(num_boxes - 1):
W
WenmuZhou 已提交
134 135
        if abs(_boxes[i + 1][0][1] - _boxes[i][0][1]) < 10 and \
                (_boxes[i + 1][0][0] < _boxes[i][0][0]):
136 137 138 139 140 141
            tmp = _boxes[i]
            _boxes[i] = _boxes[i + 1]
            _boxes[i + 1] = tmp
    return _boxes


142
def main(args):
L
LDOUBLEV 已提交
143
    image_file_list = get_image_file_list(args.image_dir)
L
LDOUBLEV 已提交
144
    text_sys = TextSystem(args)
L
LDOUBLEV 已提交
145
    is_visualize = True
W
WenmuZhou 已提交
146
    font_path = args.vis_font_path
W
WenmuZhou 已提交
147
    drop_score = args.drop_score
L
LDOUBLEV 已提交
148
    for image_file in image_file_list:
L
LDOUBLEV 已提交
149 150 151
        img, flag = check_and_read_gif(image_file)
        if not flag:
            img = cv2.imread(image_file)
L
LDOUBLEV 已提交
152 153 154 155 156 157
        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
W
WenmuZhou 已提交
158
        logger.info("Predict time of %s: %.3fs" % (image_file, elapse))
L
LDOUBLEV 已提交
159

W
WenmuZhou 已提交
160 161
        for text, score in rec_res:
            logger.info("{}, {:.3f}".format(text, score))
L
LDOUBLEV 已提交
162 163 164 165 166 167 168

        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))]

W
WenmuZhou 已提交
169 170 171 172 173 174 175
            draw_img = draw_ocr_box_txt(
                image,
                boxes,
                txts,
                scores,
                drop_score=drop_score,
                font_path=font_path)
176
            draw_img_save = "./inference_results/"
L
LDOUBLEV 已提交
177 178 179 180
            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)),
D
dyning 已提交
181
                draw_img[:, :, ::-1])
W
WenmuZhou 已提交
182
            logger.info("The visualized image saved in {}".format(
183
                os.path.join(draw_img_save, os.path.basename(image_file))))
184 185 186


if __name__ == "__main__":
L
LDOUBLEV 已提交
187
    main(utility.parse_args())