utility.py 16 KB
Newer Older
LDOUBLEV's avatar
LDOUBLEV committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
# 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 argparse
WenmuZhou's avatar
WenmuZhou committed
16
import os
WenmuZhou's avatar
WenmuZhou committed
17
import sys
LDOUBLEV's avatar
LDOUBLEV committed
18
19
import cv2
import numpy as np
LDOUBLEV's avatar
LDOUBLEV committed
20
21
import json
from PIL import Image, ImageDraw, ImageFont
22
import math
WenmuZhou's avatar
WenmuZhou committed
23
from paddle import inference
24
25
26
import time
from ppocr.utils.logging import get_logger
logger = get_logger()
LDOUBLEV's avatar
LDOUBLEV committed
27
28
29
30
31
32
33


def parse_args():
    def str2bool(v):
        return v.lower() in ("true", "t", "1")

    parser = argparse.ArgumentParser()
WenmuZhou's avatar
WenmuZhou committed
34
    # params for prediction engine
LDOUBLEV's avatar
LDOUBLEV committed
35
36
37
    parser.add_argument("--use_gpu", type=str2bool, default=True)
    parser.add_argument("--ir_optim", type=str2bool, default=True)
    parser.add_argument("--use_tensorrt", type=str2bool, default=False)
LDOUBLEV's avatar
LDOUBLEV committed
38
    parser.add_argument("--use_fp16", type=str2bool, default=False)
39
    parser.add_argument("--gpu_mem", type=int, default=500)
LDOUBLEV's avatar
LDOUBLEV committed
40

WenmuZhou's avatar
WenmuZhou committed
41
    # params for text detector
LDOUBLEV's avatar
LDOUBLEV committed
42
43
44
    parser.add_argument("--image_dir", type=str)
    parser.add_argument("--det_algorithm", type=str, default='DB')
    parser.add_argument("--det_model_dir", type=str)
WenmuZhou's avatar
WenmuZhou committed
45
46
    parser.add_argument("--det_limit_side_len", type=float, default=960)
    parser.add_argument("--det_limit_type", type=str, default='max')
LDOUBLEV's avatar
LDOUBLEV committed
47

WenmuZhou's avatar
WenmuZhou committed
48
    # DB parmas
LDOUBLEV's avatar
LDOUBLEV committed
49
50
    parser.add_argument("--det_db_thresh", type=float, default=0.3)
    parser.add_argument("--det_db_box_thresh", type=float, default=0.5)
WenmuZhou's avatar
WenmuZhou committed
51
    parser.add_argument("--det_db_unclip_ratio", type=float, default=1.6)
LDOUBLEV's avatar
LDOUBLEV committed
52
    parser.add_argument("--max_batch_size", type=int, default=10)
LDOUBLEV's avatar
LDOUBLEV committed
53
    parser.add_argument("--use_dilation", type=bool, default=False)
littletomatodonkey's avatar
littletomatodonkey committed
54
    parser.add_argument("--det_db_score_mode", type=str, default="fast")
WenmuZhou's avatar
WenmuZhou committed
55
    # EAST parmas
LDOUBLEV's avatar
LDOUBLEV committed
56
57
58
59
    parser.add_argument("--det_east_score_thresh", type=float, default=0.8)
    parser.add_argument("--det_east_cover_thresh", type=float, default=0.1)
    parser.add_argument("--det_east_nms_thresh", type=float, default=0.2)

WenmuZhou's avatar
WenmuZhou committed
60
    # SAST parmas
licx's avatar
licx committed
61
62
    parser.add_argument("--det_sast_score_thresh", type=float, default=0.5)
    parser.add_argument("--det_sast_nms_thresh", type=float, default=0.2)
63
    parser.add_argument("--det_sast_polygon", type=bool, default=False)
licx's avatar
licx committed
64

WenmuZhou's avatar
WenmuZhou committed
65
    # params for text recognizer
LDOUBLEV's avatar
LDOUBLEV committed
66
67
    parser.add_argument("--rec_algorithm", type=str, default='CRNN')
    parser.add_argument("--rec_model_dir", type=str)
tink2123's avatar
fix bug  
tink2123 committed
68
69
    parser.add_argument("--rec_image_shape", type=str, default="3, 32, 320")
    parser.add_argument("--rec_char_type", type=str, default='ch')
70
    parser.add_argument("--rec_batch_num", type=int, default=6)
tink2123's avatar
fix bug  
tink2123 committed
71
    parser.add_argument("--max_text_length", type=int, default=25)
LDOUBLEV's avatar
LDOUBLEV committed
72
73
74
75
    parser.add_argument(
        "--rec_char_dict_path",
        type=str,
        default="./ppocr/utils/ppocr_keys_v1.txt")
WenmuZhou's avatar
WenmuZhou committed
76
77
    parser.add_argument("--use_space_char", type=str2bool, default=True)
    parser.add_argument(
tink2123's avatar
tink2123 committed
78
        "--vis_font_path", type=str, default="./doc/fonts/simfang.ttf")
WenmuZhou's avatar
WenmuZhou committed
79
    parser.add_argument("--drop_score", type=float, default=0.5)
WenmuZhou's avatar
WenmuZhou committed
80

Jethong's avatar
Jethong committed
81
82
83
84
85
86
87
88
89
    # params for e2e
    parser.add_argument("--e2e_algorithm", type=str, default='PGNet')
    parser.add_argument("--e2e_model_dir", type=str)
    parser.add_argument("--e2e_limit_side_len", type=float, default=768)
    parser.add_argument("--e2e_limit_type", type=str, default='max')

    # PGNet parmas
    parser.add_argument("--e2e_pgnet_score_thresh", type=float, default=0.5)
    parser.add_argument(
Jethong's avatar
Jethong committed
90
        "--e2e_char_dict_path", type=str, default="./ppocr/utils/ic15_dict.txt")
Jethong's avatar
Jethong committed
91
    parser.add_argument("--e2e_pgnet_valid_set", type=str, default='totaltext')
Jethong's avatar
Jethong committed
92
    parser.add_argument("--e2e_pgnet_polygon", type=bool, default=True)
Jethong's avatar
Jethong committed
93
    parser.add_argument("--e2e_pgnet_mode", type=str, default='fast')
Jethong's avatar
Jethong committed
94

WenmuZhou's avatar
WenmuZhou committed
95
96
97
98
99
    # params for text classifier
    parser.add_argument("--use_angle_cls", type=str2bool, default=False)
    parser.add_argument("--cls_model_dir", type=str)
    parser.add_argument("--cls_image_shape", type=str, default="3, 48, 192")
    parser.add_argument("--label_list", type=list, default=['0', '180'])
100
    parser.add_argument("--cls_batch_num", type=int, default=6)
WenmuZhou's avatar
WenmuZhou committed
101
102
103
    parser.add_argument("--cls_thresh", type=float, default=0.9)

    parser.add_argument("--enable_mkldnn", type=str2bool, default=False)
104
    parser.add_argument("--cpu_threads", type=int, default=10)
WenmuZhou's avatar
WenmuZhou committed
105
106
    parser.add_argument("--use_pdserving", type=str2bool, default=False)

littletomatodonkey's avatar
littletomatodonkey committed
107
    parser.add_argument("--use_mp", type=str2bool, default=False)
108
109
110
    parser.add_argument("--total_process_num", type=int, default=1)
    parser.add_argument("--process_id", type=int, default=0)

LDOUBLEV's avatar
LDOUBLEV committed
111
112
113
    return parser.parse_args()


WenmuZhou's avatar
WenmuZhou committed
114
115
116
117
118
def create_predictor(args, mode, logger):
    if mode == "det":
        model_dir = args.det_model_dir
    elif mode == 'cls':
        model_dir = args.cls_model_dir
Jethong's avatar
Jethong committed
119
    elif mode == 'rec':
WenmuZhou's avatar
WenmuZhou committed
120
        model_dir = args.rec_model_dir
Jethong's avatar
Jethong committed
121
122
    else:
        model_dir = args.e2e_model_dir
WenmuZhou's avatar
WenmuZhou committed
123
124
125
126

    if model_dir is None:
        logger.info("not find {} model file path {}".format(mode, model_dir))
        sys.exit(0)
WenmuZhou's avatar
WenmuZhou committed
127
128
    model_file_path = model_dir + "/inference.pdmodel"
    params_file_path = model_dir + "/inference.pdiparams"
WenmuZhou's avatar
WenmuZhou committed
129
130
131
132
133
134
135
    if not os.path.exists(model_file_path):
        logger.info("not find model file path {}".format(model_file_path))
        sys.exit(0)
    if not os.path.exists(params_file_path):
        logger.info("not find params file path {}".format(params_file_path))
        sys.exit(0)

WenmuZhou's avatar
WenmuZhou committed
136
    config = inference.Config(model_file_path, params_file_path)
WenmuZhou's avatar
WenmuZhou committed
137
138
139

    if args.use_gpu:
        config.enable_use_gpu(args.gpu_mem, 0)
LDOUBLEV's avatar
LDOUBLEV committed
140
141
        if args.use_tensorrt:
            config.enable_tensorrt_engine(
WenmuZhou's avatar
WenmuZhou committed
142
143
                precision_mode=inference.PrecisionType.Half
                if args.use_fp16 else inference.PrecisionType.Float32,
LDOUBLEV's avatar
LDOUBLEV committed
144
                max_batch_size=args.max_batch_size)
WenmuZhou's avatar
WenmuZhou committed
145
146
    else:
        config.disable_gpu()
147
148
149
150
151
        if hasattr(args, "cpu_threads"):
            config.set_cpu_math_library_num_threads(args.cpu_threads)
        else:
            config.set_cpu_math_library_num_threads(
                10)  # default cpu threads as 10
WenmuZhou's avatar
WenmuZhou committed
152
153
154
155
156
        if args.enable_mkldnn:
            # cache 10 different shapes for mkldnn to avoid memory leak
            config.set_mkldnn_cache_capacity(10)
            config.enable_mkldnn()

LDOUBLEV's avatar
LDOUBLEV committed
157
158
    # enable memory optim
    config.enable_memory_optim()
WenmuZhou's avatar
WenmuZhou committed
159
160
    config.disable_glog_info()

WenmuZhou's avatar
WenmuZhou committed
161
162
    config.delete_pass("conv_transpose_eltwiseadd_bn_fuse_pass")
    config.switch_use_feed_fetch_ops(False)
WenmuZhou's avatar
WenmuZhou committed
163

WenmuZhou's avatar
WenmuZhou committed
164
165
    # create predictor
    predictor = inference.create_predictor(config)
WenmuZhou's avatar
WenmuZhou committed
166
167
    input_names = predictor.get_input_names()
    for name in input_names:
WenmuZhou's avatar
WenmuZhou committed
168
        input_tensor = predictor.get_input_handle(name)
WenmuZhou's avatar
WenmuZhou committed
169
170
171
    output_names = predictor.get_output_names()
    output_tensors = []
    for output_name in output_names:
WenmuZhou's avatar
WenmuZhou committed
172
        output_tensor = predictor.get_output_handle(output_name)
WenmuZhou's avatar
WenmuZhou committed
173
174
175
176
        output_tensors.append(output_tensor)
    return predictor, input_tensor, output_tensors


Jethong's avatar
Jethong committed
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
def draw_e2e_res(dt_boxes, strs, img_path):
    src_im = cv2.imread(img_path)
    for box, str in zip(dt_boxes, strs):
        box = box.astype(np.int32).reshape((-1, 1, 2))
        cv2.polylines(src_im, [box], True, color=(255, 255, 0), thickness=2)
        cv2.putText(
            src_im,
            str,
            org=(int(box[0, 0, 0]), int(box[0, 0, 1])),
            fontFace=cv2.FONT_HERSHEY_COMPLEX,
            fontScale=0.7,
            color=(0, 255, 0),
            thickness=1)
    return src_im


LDOUBLEV's avatar
LDOUBLEV committed
193
def draw_text_det_res(dt_boxes, img_path):
LDOUBLEV's avatar
LDOUBLEV committed
194
195
196
197
    src_im = cv2.imread(img_path)
    for box in dt_boxes:
        box = np.array(box).astype(np.int32).reshape(-1, 2)
        cv2.polylines(src_im, [box], True, color=(255, 255, 0), thickness=2)
LDOUBLEV's avatar
LDOUBLEV committed
198
    return src_im
LDOUBLEV's avatar
LDOUBLEV committed
199
200


LDOUBLEV's avatar
LDOUBLEV committed
201
202
def resize_img(img, input_size=600):
    """
LDOUBLEV's avatar
LDOUBLEV committed
203
    resize img and limit the longest side of the image to input_size
LDOUBLEV's avatar
LDOUBLEV committed
204
205
206
207
208
    """
    img = np.array(img)
    im_shape = img.shape
    im_size_max = np.max(im_shape[0:2])
    im_scale = float(input_size) / float(im_size_max)
WenmuZhou's avatar
WenmuZhou committed
209
210
    img = cv2.resize(img, None, None, fx=im_scale, fy=im_scale)
    return img
LDOUBLEV's avatar
LDOUBLEV committed
211
212


WenmuZhou's avatar
WenmuZhou committed
213
214
215
216
217
218
def draw_ocr(image,
             boxes,
             txts=None,
             scores=None,
             drop_score=0.5,
             font_path="./doc/simfang.ttf"):
219
220
221
    """
    Visualize the results of OCR detection and recognition
    args:
LDOUBLEV's avatar
LDOUBLEV committed
222
        image(Image|array): RGB image
223
224
225
226
        boxes(list): boxes with shape(N, 4, 2)
        txts(list): the texts
        scores(list): txxs corresponding scores
        drop_score(float): only scores greater than drop_threshold will be visualized
WenmuZhou's avatar
WenmuZhou committed
227
        font_path: the path of font which is used to draw text
228
229
230
    return(array):
        the visualized img
    """
LDOUBLEV's avatar
LDOUBLEV committed
231
232
    if scores is None:
        scores = [1] * len(boxes)
WenmuZhou's avatar
WenmuZhou committed
233
234
235
236
    box_num = len(boxes)
    for i in range(box_num):
        if scores is not None and (scores[i] < drop_score or
                                   math.isnan(scores[i])):
LDOUBLEV's avatar
LDOUBLEV committed
237
            continue
WenmuZhou's avatar
WenmuZhou committed
238
        box = np.reshape(np.array(boxes[i]), [-1, 1, 2]).astype(np.int64)
LDOUBLEV's avatar
LDOUBLEV committed
239
        image = cv2.polylines(np.array(image), [box], True, (255, 0, 0), 2)
WenmuZhou's avatar
WenmuZhou committed
240
    if txts is not None:
LDOUBLEV's avatar
LDOUBLEV committed
241
        img = np.array(resize_img(image, input_size=600))
242
        txt_img = text_visual(
WenmuZhou's avatar
WenmuZhou committed
243
244
245
246
247
248
            txts,
            scores,
            img_h=img.shape[0],
            img_w=600,
            threshold=drop_score,
            font_path=font_path)
249
        img = np.concatenate([np.array(img), np.array(txt_img)], axis=1)
LDOUBLEV's avatar
LDOUBLEV committed
250
251
        return img
    return image
252
253


WenmuZhou's avatar
WenmuZhou committed
254
255
256
257
258
259
def draw_ocr_box_txt(image,
                     boxes,
                     txts,
                     scores=None,
                     drop_score=0.5,
                     font_path="./doc/simfang.ttf"):
260
261
262
    h, w = image.height, image.width
    img_left = image.copy()
    img_right = Image.new('RGB', (w, h), (255, 255, 255))
263
264

    import random
LDOUBLEV's avatar
LDOUBLEV committed
265

266
267
268
    random.seed(0)
    draw_left = ImageDraw.Draw(img_left)
    draw_right = ImageDraw.Draw(img_right)
WenmuZhou's avatar
WenmuZhou committed
269
270
271
    for idx, (box, txt) in enumerate(zip(boxes, txts)):
        if scores is not None and scores[idx] < drop_score:
            continue
tink2123's avatar
tink2123 committed
272
273
        color = (random.randint(0, 255), random.randint(0, 255),
                 random.randint(0, 255))
274
        draw_left.polygon(box, fill=color)
tink2123's avatar
tink2123 committed
275
276
277
278
279
280
281
282
283
284
        draw_right.polygon(
            [
                box[0][0], box[0][1], box[1][0], box[1][1], box[2][0],
                box[2][1], box[3][0], box[3][1]
            ],
            outline=color)
        box_height = math.sqrt((box[0][0] - box[3][0])**2 + (box[0][1] - box[3][
            1])**2)
        box_width = math.sqrt((box[0][0] - box[1][0])**2 + (box[0][1] - box[1][
            1])**2)
285
286
        if box_height > 2 * box_width:
            font_size = max(int(box_width * 0.9), 10)
WenmuZhou's avatar
WenmuZhou committed
287
            font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
288
289
290
            cur_y = box[0][1]
            for c in txt:
                char_size = font.getsize(c)
tink2123's avatar
tink2123 committed
291
292
                draw_right.text(
                    (box[0][0] + 3, cur_y), c, fill=(0, 0, 0), font=font)
293
294
295
                cur_y += char_size[1]
        else:
            font_size = max(int(box_height * 0.8), 10)
WenmuZhou's avatar
WenmuZhou committed
296
            font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
tink2123's avatar
tink2123 committed
297
298
            draw_right.text(
                [box[0][0], box[0][1]], txt, fill=(0, 0, 0), font=font)
299
300
301
302
    img_left = Image.blend(image, img_left, 0.5)
    img_show = Image.new('RGB', (w * 2, h), (255, 255, 255))
    img_show.paste(img_left, (0, 0, w, h))
    img_show.paste(img_right, (w, 0, w * 2, h))
303
304
305
    return np.array(img_show)


306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
def str_count(s):
    """
    Count the number of Chinese characters,
    a single English character and a single number
    equal to half the length of Chinese characters.
    args:
        s(string): the input of string
    return(int):
        the number of Chinese characters
    """
    import string
    count_zh = count_pu = 0
    s_len = len(s)
    en_dg_count = 0
    for c in s:
        if c in string.ascii_letters or c.isdigit() or c.isspace():
            en_dg_count += 1
        elif c.isalpha():
            count_zh += 1
        else:
            count_pu += 1
    return s_len - math.ceil(en_dg_count / 2)


WenmuZhou's avatar
WenmuZhou committed
330
331
332
333
334
335
def text_visual(texts,
                scores,
                img_h=400,
                img_w=600,
                threshold=0.,
                font_path="./doc/simfang.ttf"):
336
337
338
339
340
341
342
    """
    create new blank img and draw txt on it
    args:
        texts(list): the text will be draw
        scores(list|None): corresponding score of each txt
        img_h(int): the height of blank img
        img_w(int): the width of blank img
WenmuZhou's avatar
WenmuZhou committed
343
        font_path: the path of font which is used to draw text
344
345
346
347
348
349
350
351
352
    return(array):
    """
    if scores is not None:
        assert len(texts) == len(
            scores), "The number of txts and corresponding scores must match"

    def create_blank_img():
        blank_img = np.ones(shape=[img_h, img_w], dtype=np.int8) * 255
        blank_img[:, img_w - 1:] = 0
LDOUBLEV's avatar
LDOUBLEV committed
353
354
        blank_img = Image.fromarray(blank_img).convert("RGB")
        draw_txt = ImageDraw.Draw(blank_img)
355
        return blank_img, draw_txt
LDOUBLEV's avatar
LDOUBLEV committed
356

357
358
359
360
    blank_img, draw_txt = create_blank_img()

    font_size = 20
    txt_color = (0, 0, 0)
WenmuZhou's avatar
WenmuZhou committed
361
    font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
362
363
364

    gap = font_size + 5
    txt_img_list = []
LDOUBLEV's avatar
LDOUBLEV committed
365
    count, index = 1, 0
366
367
    for idx, txt in enumerate(texts):
        index += 1
LDOUBLEV's avatar
LDOUBLEV committed
368
        if scores[idx] < threshold or math.isnan(scores[idx]):
369
370
371
372
373
374
375
376
377
378
379
            index -= 1
            continue
        first_line = True
        while str_count(txt) >= img_w // font_size - 4:
            tmp = txt
            txt = tmp[:img_w // font_size - 4]
            if first_line:
                new_txt = str(index) + ': ' + txt
                first_line = False
            else:
                new_txt = '    ' + txt
LDOUBLEV's avatar
LDOUBLEV committed
380
            draw_txt.text((0, gap * count), new_txt, txt_color, font=font)
381
382
383
384
385
            txt = tmp[img_w // font_size - 4:]
            if count >= img_h // gap - 1:
                txt_img_list.append(np.array(blank_img))
                blank_img, draw_txt = create_blank_img()
                count = 0
LDOUBLEV's avatar
LDOUBLEV committed
386
            count += 1
387
388
389
        if first_line:
            new_txt = str(index) + ': ' + txt + '   ' + '%.3f' % (scores[idx])
        else:
LDOUBLEV's avatar
LDOUBLEV committed
390
            new_txt = "  " + txt + "  " + '%.3f' % (scores[idx])
LDOUBLEV's avatar
LDOUBLEV committed
391
        draw_txt.text((0, gap * count), new_txt, txt_color, font=font)
392
        # whether add new blank img or not
LDOUBLEV's avatar
LDOUBLEV committed
393
        if count >= img_h // gap - 1 and idx + 1 < len(texts):
394
395
396
            txt_img_list.append(np.array(blank_img))
            blank_img, draw_txt = create_blank_img()
            count = 0
LDOUBLEV's avatar
LDOUBLEV committed
397
        count += 1
398
399
400
401
402
403
    txt_img_list.append(np.array(blank_img))
    if len(txt_img_list) == 1:
        blank_img = np.array(txt_img_list[0])
    else:
        blank_img = np.concatenate(txt_img_list, axis=1)
    return np.array(blank_img)
LDOUBLEV's avatar
LDOUBLEV committed
404
405


dyning's avatar
dyning committed
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
def base64_to_cv2(b64str):
    import base64
    data = base64.b64decode(b64str.encode('utf8'))
    data = np.fromstring(data, np.uint8)
    data = cv2.imdecode(data, cv2.IMREAD_COLOR)
    return data


def draw_boxes(image, boxes, scores=None, drop_score=0.5):
    if scores is None:
        scores = [1] * len(boxes)
    for (box, score) in zip(boxes, scores):
        if score < drop_score:
            continue
        box = np.reshape(np.array(box), [-1, 1, 2]).astype(np.int64)
        image = cv2.polylines(np.array(image), [box], True, (255, 0, 0), 2)
    return image


LDOUBLEV's avatar
LDOUBLEV committed
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
if __name__ == '__main__':
    test_img = "./doc/test_v2"
    predict_txt = "./doc/predict.txt"
    f = open(predict_txt, 'r')
    data = f.readlines()
    img_path, anno = data[0].strip().split('\t')
    img_name = os.path.basename(img_path)
    img_path = os.path.join(test_img, img_name)
    image = Image.open(img_path)

    data = json.loads(anno)
    boxes, txts, scores = [], [], []
    for dic in data:
        boxes.append(dic['points'])
        txts.append(dic['transcription'])
        scores.append(round(dic['scores'], 3))

WenmuZhou's avatar
WenmuZhou committed
442
    new_img = draw_ocr(image, boxes, txts, scores)
LDOUBLEV's avatar
LDOUBLEV committed
443

MissPenguin's avatar
MissPenguin committed
444
    cv2.imwrite(img_name, new_img)