predict_det.py 9.5 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

L
LDOUBLEV 已提交
34
# import tools.infer.benchmark_utils as benchmark_utils
L
LDOUBLEV 已提交
35

W
WenmuZhou 已提交
36
logger = get_logger()
L
LDOUBLEV 已提交
37
import auto_log
W
WenmuZhou 已提交
38

L
LDOUBLEV 已提交
39 40 41

class TextDetector(object):
    def __init__(self, args):
L
LDOUBLEV 已提交
42
        self.args = args
L
LDOUBLEV 已提交
43
        self.det_algorithm = args.det_algorithm
M
MissPenguin 已提交
44
        pre_process_list = [{
45 46
            'DetResizeForTest': {
                'limit_side_len': args.det_limit_side_len,
W
WenmuZhou 已提交
47
                'limit_type': args.det_limit_type,
48
            }
M
MissPenguin 已提交
49 50 51 52 53 54 55 56 57 58 59 60 61 62
        }, {
            '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
63 64
        postprocess_params = {}
        if self.det_algorithm == "DB":
W
WenmuZhou 已提交
65
            postprocess_params['name'] = 'DBPostProcess'
L
LDOUBLEV 已提交
66 67 68
            postprocess_params["thresh"] = args.det_db_thresh
            postprocess_params["box_thresh"] = args.det_db_box_thresh
            postprocess_params["max_candidates"] = 1000
69
            postprocess_params["unclip_ratio"] = args.det_db_unclip_ratio
L
LDOUBLEV 已提交
70
            postprocess_params["use_dilation"] = args.use_dilation
littletomatodonkey's avatar
littletomatodonkey 已提交
71
            postprocess_params["score_mode"] = args.det_db_score_mode
M
MissPenguin 已提交
72
        elif self.det_algorithm == "EAST":
W
WenmuZhou 已提交
73
            postprocess_params['name'] = 'EASTPostProcess'
M
MissPenguin 已提交
74 75 76 77
            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
78
            pre_process_list[0] = {
W
WenmuZhou 已提交
79 80 81
                'DetResizeForTest': {
                    'resize_long': args.det_limit_side_len
                }
M
MissPenguin 已提交
82
            }
W
WenmuZhou 已提交
83
            postprocess_params['name'] = 'SASTPostProcess'
M
MissPenguin 已提交
84 85 86 87 88 89 90 91 92 93 94
            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
95 96 97 98
        else:
            logger.info("unknown det_algorithm:{}".format(self.det_algorithm))
            sys.exit(0)

W
WenmuZhou 已提交
99 100
        self.preprocess_op = create_operators(pre_process_list)
        self.postprocess_op = build_post_process(postprocess_params)
L
LDOUBLEV 已提交
101 102 103
        self.predictor, self.input_tensor, self.output_tensors, self.config = utility.create_predictor(
            args, 'det', logger)

L
LDOUBLEV 已提交
104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119
        pid = os.getpid()
        self.autolog = auto_log.AutoLogger(
            model_name="det",
            model_precision="fp32",
            batch_size=1,
            data_shape="dynamic",
            save_path="./output/auto_log.lpg",
            inference_config=self.config,
            pids=pid,
            process_name=None,
            gpu_ids=0,
            time_keys=[
                'preprocess_time', 'inference_time', 'postprocess_time'
            ],
            warmup=10)

L
LDOUBLEV 已提交
120
    def order_points_clockwise(self, pts):
121 122
        """
        reference from: https://github.com/jrosebr1/imutils/blob/master/imutils/perspective.py
L
LDOUBLEV 已提交
123
        # sort the points based on their x-coordinates
124
        """
L
LDOUBLEV 已提交
125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143
        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
144
    def clip_det_res(self, points, img_height, img_width):
145
        for pno in range(points.shape[0]):
D
dyning 已提交
146 147
            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
148 149 150 151 152 153 154
        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
155
            box = self.clip_det_res(box, img_height, img_width)
L
LDOUBLEV 已提交
156 157
            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
158
            if rect_width <= 3 or rect_height <= 3:
L
LDOUBLEV 已提交
159 160 161 162 163
                continue
            dt_boxes_new.append(box)
        dt_boxes = np.array(dt_boxes_new)
        return dt_boxes

164 165 166 167 168 169 170 171
    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
172

L
LDOUBLEV 已提交
173 174
    def __call__(self, img):
        ori_im = img.copy()
W
WenmuZhou 已提交
175
        data = {'image': img}
L
LDOUBLEV 已提交
176 177

        st = time.time()
L
LDOUBLEV 已提交
178 179 180

        self.autolog.times.start()

W
WenmuZhou 已提交
181 182 183
        data = transform(data, self.preprocess_op)
        img, shape_list = data
        if img is None:
L
LDOUBLEV 已提交
184
            return None, 0
W
WenmuZhou 已提交
185 186
        img = np.expand_dims(img, axis=0)
        shape_list = np.expand_dims(shape_list, axis=0)
187
        img = img.copy()
L
LDOUBLEV 已提交
188

L
LDOUBLEV 已提交
189 190
        self.autolog.times.stamp()

W
WenmuZhou 已提交
191 192
        self.input_tensor.copy_from_cpu(img)
        self.predictor.run()
193 194 195 196 197
        outputs = []
        for output_tensor in self.output_tensors:
            output = output_tensor.copy_to_cpu()
            outputs.append(output)

L
LDOUBLEV 已提交
198 199
        self.autolog.times.stamp()

M
MissPenguin 已提交
200 201 202 203 204 205 206 207 208
        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
209
        elif self.det_algorithm == 'DB':
W
WenmuZhou 已提交
210
            preds['maps'] = outputs[0]
W
WenmuZhou 已提交
211 212
        else:
            raise NotImplementedError
L
LDOUBLEV 已提交
213

W
fix mem  
WenmuZhou 已提交
214
        self.predictor.try_shrink_memory()
W
WenmuZhou 已提交
215 216
        post_result = self.postprocess_op(preds, shape_list)
        dt_boxes = post_result[0]['points']
M
MissPenguin 已提交
217 218 219 220
        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
221

L
LDOUBLEV 已提交
222
        self.autolog.times.end(stamp=True)
L
LDOUBLEV 已提交
223
        et = time.time()
L
LDOUBLEV 已提交
224
        self.autolog.get_avg_mem_mb()
L
LDOUBLEV 已提交
225
        return dt_boxes, et - st
L
LDOUBLEV 已提交
226 227 228 229


if __name__ == "__main__":
    args = utility.parse_args()
L
LDOUBLEV 已提交
230
    image_file_list = get_image_file_list(args.image_dir)
L
LDOUBLEV 已提交
231 232 233
    text_detector = TextDetector(args)
    count = 0
    total_time = 0
littletomatodonkey's avatar
littletomatodonkey 已提交
234
    draw_img_save = "./inference_results"
L
LDOUBLEV 已提交
235

L
LDOUBLEV 已提交
236 237 238 239 240
    if args.warmup:
        img = np.random.uniform(0, 255, [640, 640, 3]).astype(np.uint8)
        for i in range(10):
            res = text_detector(img)

L
LDOUBLEV 已提交
241
    text_detector.autolog.mem_info._start_subprocess()
L
LDOUBLEV 已提交
242

littletomatodonkey's avatar
littletomatodonkey 已提交
243 244
    if not os.path.exists(draw_img_save):
        os.makedirs(draw_img_save)
L
LDOUBLEV 已提交
245
    for image_file in image_file_list:
L
LDOUBLEV 已提交
246 247 248
        img, flag = check_and_read_gif(image_file)
        if not flag:
            img = cv2.imread(image_file)
L
LDOUBLEV 已提交
249 250 251
        if img is None:
            logger.info("error in loading image:{}".format(image_file))
            continue
L
LDOUBLEV 已提交
252 253 254
        st = time.time()
        dt_boxes, _ = text_detector(img)
        elapse = time.time() - st
L
LDOUBLEV 已提交
255 256 257
        if count > 0:
            total_time += elapse
        count += 1
L
LDOUBLEV 已提交
258

W
WenmuZhou 已提交
259
        logger.info("Predict time of {}: {}".format(image_file, elapse))
D
dyning 已提交
260
        src_im = utility.draw_text_det_res(dt_boxes, image_file)
W
WenmuZhou 已提交
261
        img_name_pure = os.path.split(image_file)[-1]
W
WenmuZhou 已提交
262 263
        img_path = os.path.join(draw_img_save,
                                "det_res_{}".format(img_name_pure))
L
LDOUBLEV 已提交
264

W
WenmuZhou 已提交
265
        logger.info("The visualized image saved in {}".format(img_path))
L
LDOUBLEV 已提交
266 267

    text_detector.autolog.report()