predict_det.py 8.5 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.
L
LDOUBLEV 已提交
14 15
import os
import sys
W
WenmuZhou 已提交
16

17
__dir__ = os.path.dirname(os.path.abspath(__file__))
L
LDOUBLEV 已提交
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'

23 24 25 26 27
import cv2
import numpy as np
import time
import sys

L
LDOUBLEV 已提交
28
import tools.infer.utility as utility
W
WenmuZhou 已提交
29
from ppocr.utils.logging import get_logger
L
LDOUBLEV 已提交
30
from ppocr.utils.utility import get_image_file_list, check_and_read_gif
W
WenmuZhou 已提交
31 32
from ppocr.data import create_operators, transform
from ppocr.postprocess import build_post_process
L
LDOUBLEV 已提交
33

W
WenmuZhou 已提交
34 35
logger = get_logger()

L
LDOUBLEV 已提交
36 37 38

class TextDetector(object):
    def __init__(self, args):
L
LDOUBLEV 已提交
39
        self.args = args
L
LDOUBLEV 已提交
40
        self.det_algorithm = args.det_algorithm
M
MissPenguin 已提交
41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59
        pre_process_list = [{
            'DetResizeForTest': {
                'limit_side_len': args.det_limit_side_len,
                'limit_type': args.det_limit_type
            }
        }, {
            'NormalizeImage': {
                'std': [0.229, 0.224, 0.225],
                'mean': [0.485, 0.456, 0.406],
                'scale': '1./255.',
                'order': 'hwc'
            }
        }, {
            'ToCHWImage': None
        }, {
            'KeepKeys': {
                'keep_keys': ['image', 'shape']
            }
        }]
L
LDOUBLEV 已提交
60 61
        postprocess_params = {}
        if self.det_algorithm == "DB":
W
WenmuZhou 已提交
62
            postprocess_params['name'] = 'DBPostProcess'
L
LDOUBLEV 已提交
63 64 65
            postprocess_params["thresh"] = args.det_db_thresh
            postprocess_params["box_thresh"] = args.det_db_box_thresh
            postprocess_params["max_candidates"] = 1000
66
            postprocess_params["unclip_ratio"] = args.det_db_unclip_ratio
L
LDOUBLEV 已提交
67
            postprocess_params["use_dilation"] = args.use_dilation
M
MissPenguin 已提交
68
        elif self.det_algorithm == "EAST":
W
WenmuZhou 已提交
69
            postprocess_params['name'] = 'EASTPostProcess'
M
MissPenguin 已提交
70 71 72 73
            postprocess_params["score_thresh"] = args.det_east_score_thresh
            postprocess_params["cover_thresh"] = args.det_east_cover_thresh
            postprocess_params["nms_thresh"] = args.det_east_nms_thresh
        elif self.det_algorithm == "SAST":
M
MissPenguin 已提交
74
            pre_process_list[0] = {
W
WenmuZhou 已提交
75 76 77
                'DetResizeForTest': {
                    'resize_long': args.det_limit_side_len
                }
M
MissPenguin 已提交
78
            }
W
WenmuZhou 已提交
79
            postprocess_params['name'] = 'SASTPostProcess'
M
MissPenguin 已提交
80 81 82 83 84 85 86 87 88 89 90
            postprocess_params["score_thresh"] = args.det_sast_score_thresh
            postprocess_params["nms_thresh"] = args.det_sast_nms_thresh
            self.det_sast_polygon = args.det_sast_polygon
            if self.det_sast_polygon:
                postprocess_params["sample_pts_num"] = 6
                postprocess_params["expand_scale"] = 1.2
                postprocess_params["shrink_ratio_of_width"] = 0.2
            else:
                postprocess_params["sample_pts_num"] = 2
                postprocess_params["expand_scale"] = 1.0
                postprocess_params["shrink_ratio_of_width"] = 0.3
L
LDOUBLEV 已提交
91 92 93 94
        else:
            logger.info("unknown det_algorithm:{}".format(self.det_algorithm))
            sys.exit(0)

W
WenmuZhou 已提交
95 96
        self.preprocess_op = create_operators(pre_process_list)
        self.postprocess_op = build_post_process(postprocess_params)
97 98 99
        self.predictor, self.input_tensor, self.output_tensors = utility.create_predictor(
            args, 'det', logger)  # paddle.jit.load(args.det_model_dir)
        # self.predictor.eval()
L
LDOUBLEV 已提交
100 101

    def order_points_clockwise(self, pts):
102 103
        """
        reference from: https://github.com/jrosebr1/imutils/blob/master/imutils/perspective.py
L
LDOUBLEV 已提交
104
        # sort the points based on their x-coordinates
105
        """
L
LDOUBLEV 已提交
106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124
        xSorted = pts[np.argsort(pts[:, 0]), :]

        # grab the left-most and right-most points from the sorted
        # x-roodinate points
        leftMost = xSorted[:2, :]
        rightMost = xSorted[2:, :]

        # now, sort the left-most coordinates according to their
        # y-coordinates so we can grab the top-left and bottom-left
        # points, respectively
        leftMost = leftMost[np.argsort(leftMost[:, 1]), :]
        (tl, bl) = leftMost

        rightMost = rightMost[np.argsort(rightMost[:, 1]), :]
        (tr, br) = rightMost

        rect = np.array([tl, tr, br, bl], dtype="float32")
        return rect

D
dyning 已提交
125
    def clip_det_res(self, points, img_height, img_width):
126
        for pno in range(points.shape[0]):
D
dyning 已提交
127 128
            points[pno, 0] = int(min(max(points[pno, 0], 0), img_width - 1))
            points[pno, 1] = int(min(max(points[pno, 1], 0), img_height - 1))
L
LDOUBLEV 已提交
129 130 131 132 133 134 135
        return points

    def filter_tag_det_res(self, dt_boxes, image_shape):
        img_height, img_width = image_shape[0:2]
        dt_boxes_new = []
        for box in dt_boxes:
            box = self.order_points_clockwise(box)
D
dyning 已提交
136
            box = self.clip_det_res(box, img_height, img_width)
L
LDOUBLEV 已提交
137 138
            rect_width = int(np.linalg.norm(box[0] - box[1]))
            rect_height = int(np.linalg.norm(box[0] - box[3]))
M
MissPenguin 已提交
139
            if rect_width <= 3 or rect_height <= 3:
L
LDOUBLEV 已提交
140 141 142 143 144
                continue
            dt_boxes_new.append(box)
        dt_boxes = np.array(dt_boxes_new)
        return dt_boxes

145 146 147 148 149 150 151 152
    def filter_tag_det_res_only_clip(self, dt_boxes, image_shape):
        img_height, img_width = image_shape[0:2]
        dt_boxes_new = []
        for box in dt_boxes:
            box = self.clip_det_res(box, img_height, img_width)
            dt_boxes_new.append(box)
        dt_boxes = np.array(dt_boxes_new)
        return dt_boxes
153

L
LDOUBLEV 已提交
154 155
    def __call__(self, img):
        ori_im = img.copy()
W
WenmuZhou 已提交
156 157 158 159
        data = {'image': img}
        data = transform(data, self.preprocess_op)
        img, shape_list = data
        if img is None:
L
LDOUBLEV 已提交
160
            return None, 0
W
WenmuZhou 已提交
161 162
        img = np.expand_dims(img, axis=0)
        shape_list = np.expand_dims(shape_list, axis=0)
163
        img = img.copy()
L
LDOUBLEV 已提交
164
        starttime = time.time()
165

W
WenmuZhou 已提交
166 167
        self.input_tensor.copy_from_cpu(img)
        self.predictor.run()
168 169 170 171 172
        outputs = []
        for output_tensor in self.output_tensors:
            output = output_tensor.copy_to_cpu()
            outputs.append(output)

M
MissPenguin 已提交
173 174 175 176 177 178 179 180 181
        preds = {}
        if self.det_algorithm == "EAST":
            preds['f_geo'] = outputs[0]
            preds['f_score'] = outputs[1]
        elif self.det_algorithm == 'SAST':
            preds['f_border'] = outputs[0]
            preds['f_score'] = outputs[1]
            preds['f_tco'] = outputs[2]
            preds['f_tvo'] = outputs[3]
W
WenmuZhou 已提交
182
        elif self.det_algorithm == 'DB':
W
WenmuZhou 已提交
183
            preds['maps'] = outputs[0]
W
WenmuZhou 已提交
184 185
        else:
            raise NotImplementedError
M
MissPenguin 已提交
186

W
WenmuZhou 已提交
187 188
        post_result = self.postprocess_op(preds, shape_list)
        dt_boxes = post_result[0]['points']
M
MissPenguin 已提交
189 190 191 192
        if self.det_algorithm == "SAST" and self.det_sast_polygon:
            dt_boxes = self.filter_tag_det_res_only_clip(dt_boxes, ori_im.shape)
        else:
            dt_boxes = self.filter_tag_det_res(dt_boxes, ori_im.shape)
L
LDOUBLEV 已提交
193 194 195 196 197 198
        elapse = time.time() - starttime
        return dt_boxes, elapse


if __name__ == "__main__":
    args = utility.parse_args()
L
LDOUBLEV 已提交
199
    image_file_list = get_image_file_list(args.image_dir)
L
LDOUBLEV 已提交
200 201 202
    text_detector = TextDetector(args)
    count = 0
    total_time = 0
littletomatodonkey's avatar
littletomatodonkey 已提交
203 204 205
    draw_img_save = "./inference_results"
    if not os.path.exists(draw_img_save):
        os.makedirs(draw_img_save)
L
LDOUBLEV 已提交
206
    for image_file in image_file_list:
L
LDOUBLEV 已提交
207 208 209
        img, flag = check_and_read_gif(image_file)
        if not flag:
            img = cv2.imread(image_file)
L
LDOUBLEV 已提交
210 211 212 213 214 215 216
        if img is None:
            logger.info("error in loading image:{}".format(image_file))
            continue
        dt_boxes, elapse = text_detector(img)
        if count > 0:
            total_time += elapse
        count += 1
W
WenmuZhou 已提交
217
        logger.info("Predict time of {}: {}".format(image_file, elapse))
D
dyning 已提交
218
        src_im = utility.draw_text_det_res(dt_boxes, image_file)
W
WenmuZhou 已提交
219
        img_name_pure = os.path.split(image_file)[-1]
W
WenmuZhou 已提交
220 221
        img_path = os.path.join(draw_img_save,
                                "det_res_{}".format(img_name_pure))
W
WenmuZhou 已提交
222
        cv2.imwrite(img_path, src_im)
W
WenmuZhou 已提交
223
        logger.info("The visualized image saved in {}".format(img_path))
224
    if count > 1:
225
        logger.info("Avg Time: {}".format(total_time / (count - 1)))