utility.py 15.7 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
LDOUBLEV's avatar
LDOUBLEV committed
24
25
26
27
28
29
30


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

    parser = argparse.ArgumentParser()
WenmuZhou's avatar
WenmuZhou committed
31
    # params for prediction engine
LDOUBLEV's avatar
LDOUBLEV committed
32
33
34
    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
35
    parser.add_argument("--use_fp16", type=str2bool, default=False)
36
    parser.add_argument("--gpu_mem", type=int, default=500)
LDOUBLEV's avatar
LDOUBLEV committed
37

WenmuZhou's avatar
WenmuZhou committed
38
    # params for text detector
LDOUBLEV's avatar
LDOUBLEV committed
39
40
41
    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
42
43
    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
44

WenmuZhou's avatar
WenmuZhou committed
45
    # DB parmas
LDOUBLEV's avatar
LDOUBLEV committed
46
47
    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
48
    parser.add_argument("--det_db_unclip_ratio", type=float, default=1.6)
LDOUBLEV's avatar
LDOUBLEV committed
49
    parser.add_argument("--max_batch_size", type=int, default=10)
LDOUBLEV's avatar
LDOUBLEV committed
50
    parser.add_argument("--use_dilation", type=bool, default=False)
WenmuZhou's avatar
WenmuZhou committed
51
    # EAST parmas
LDOUBLEV's avatar
LDOUBLEV committed
52
53
54
55
    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
56
    # SAST parmas
licx's avatar
licx committed
57
58
    parser.add_argument("--det_sast_score_thresh", type=float, default=0.5)
    parser.add_argument("--det_sast_nms_thresh", type=float, default=0.2)
59
    parser.add_argument("--det_sast_polygon", type=bool, default=False)
licx's avatar
licx committed
60

WenmuZhou's avatar
WenmuZhou committed
61
    # params for text recognizer
LDOUBLEV's avatar
LDOUBLEV committed
62
63
    parser.add_argument("--rec_algorithm", type=str, default='CRNN')
    parser.add_argument("--rec_model_dir", type=str)
tink2123's avatar
fix bug  
tink2123 committed
64
65
    parser.add_argument("--rec_image_shape", type=str, default="3, 32, 320")
    parser.add_argument("--rec_char_type", type=str, default='ch')
66
    parser.add_argument("--rec_batch_num", type=int, default=6)
tink2123's avatar
fix bug  
tink2123 committed
67
    parser.add_argument("--max_text_length", type=int, default=25)
LDOUBLEV's avatar
LDOUBLEV committed
68
69
70
71
    parser.add_argument(
        "--rec_char_dict_path",
        type=str,
        default="./ppocr/utils/ppocr_keys_v1.txt")
WenmuZhou's avatar
WenmuZhou committed
72
73
    parser.add_argument("--use_space_char", type=str2bool, default=True)
    parser.add_argument(
tink2123's avatar
tink2123 committed
74
        "--vis_font_path", type=str, default="./doc/fonts/simfang.ttf")
WenmuZhou's avatar
WenmuZhou committed
75
    parser.add_argument("--drop_score", type=float, default=0.5)
WenmuZhou's avatar
WenmuZhou committed
76

Jethong's avatar
Jethong committed
77
78
79
80
81
82
83
84
85
    # 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
86
        "--e2e_char_dict_path", type=str, default="./ppocr/utils/ic15_dict.txt")
Jethong's avatar
Jethong committed
87
    parser.add_argument("--e2e_pgnet_valid_set", type=str, default='totaltext')
Jethong's avatar
Jethong committed
88
    parser.add_argument("--e2e_pgnet_polygon", type=bool, default=True)
Jethong's avatar
Jethong committed
89

WenmuZhou's avatar
WenmuZhou committed
90
91
92
93
94
    # 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'])
95
    parser.add_argument("--cls_batch_num", type=int, default=6)
WenmuZhou's avatar
WenmuZhou committed
96
97
98
99
100
    parser.add_argument("--cls_thresh", type=float, default=0.9)

    parser.add_argument("--enable_mkldnn", type=str2bool, default=False)
    parser.add_argument("--use_pdserving", type=str2bool, default=False)

101
102
103
    parser.add_argument("--total_process_num", type=int, default=1)
    parser.add_argument("--process_id", type=int, default=0)

LDOUBLEV's avatar
LDOUBLEV committed
104
105
106
    return parser.parse_args()


WenmuZhou's avatar
WenmuZhou committed
107
108
109
110
111
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
112
    elif mode == 'rec':
WenmuZhou's avatar
WenmuZhou committed
113
        model_dir = args.rec_model_dir
Jethong's avatar
Jethong committed
114
115
    else:
        model_dir = args.e2e_model_dir
WenmuZhou's avatar
WenmuZhou committed
116
117
118
119

    if model_dir is None:
        logger.info("not find {} model file path {}".format(mode, model_dir))
        sys.exit(0)
WenmuZhou's avatar
WenmuZhou committed
120
121
    model_file_path = model_dir + "/inference.pdmodel"
    params_file_path = model_dir + "/inference.pdiparams"
WenmuZhou's avatar
WenmuZhou committed
122
123
124
125
126
127
128
    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
129
    config = inference.Config(model_file_path, params_file_path)
WenmuZhou's avatar
WenmuZhou committed
130
131
132

    if args.use_gpu:
        config.enable_use_gpu(args.gpu_mem, 0)
LDOUBLEV's avatar
LDOUBLEV committed
133
134
        if args.use_tensorrt:
            config.enable_tensorrt_engine(
WenmuZhou's avatar
WenmuZhou committed
135
136
                precision_mode=inference.PrecisionType.Half
                if args.use_fp16 else inference.PrecisionType.Float32,
LDOUBLEV's avatar
LDOUBLEV committed
137
                max_batch_size=args.max_batch_size)
WenmuZhou's avatar
WenmuZhou committed
138
139
140
141
142
143
144
    else:
        config.disable_gpu()
        config.set_cpu_math_library_num_threads(6)
        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
145
            #  TODO LDOUBLEV: fix mkldnn bug when bach_size  > 1
LDOUBLEV's avatar
LDOUBLEV committed
146
            #config.set_mkldnn_op({'conv2d', 'depthwise_conv2d', 'pool2d', 'batch_norm'})
147
            args.rec_batch_num = 1
WenmuZhou's avatar
WenmuZhou committed
148

LDOUBLEV's avatar
LDOUBLEV committed
149
150
    # enable memory optim
    config.enable_memory_optim()
WenmuZhou's avatar
WenmuZhou committed
151
152
    config.disable_glog_info()

WenmuZhou's avatar
WenmuZhou committed
153
154
    config.delete_pass("conv_transpose_eltwiseadd_bn_fuse_pass")
    config.switch_use_feed_fetch_ops(False)
WenmuZhou's avatar
WenmuZhou committed
155

WenmuZhou's avatar
WenmuZhou committed
156
157
    # create predictor
    predictor = inference.create_predictor(config)
WenmuZhou's avatar
WenmuZhou committed
158
159
    input_names = predictor.get_input_names()
    for name in input_names:
WenmuZhou's avatar
WenmuZhou committed
160
        input_tensor = predictor.get_input_handle(name)
WenmuZhou's avatar
WenmuZhou committed
161
162
163
    output_names = predictor.get_output_names()
    output_tensors = []
    for output_name in output_names:
WenmuZhou's avatar
WenmuZhou committed
164
        output_tensor = predictor.get_output_handle(output_name)
WenmuZhou's avatar
WenmuZhou committed
165
166
167
168
        output_tensors.append(output_tensor)
    return predictor, input_tensor, output_tensors


Jethong's avatar
Jethong committed
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
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
185
def draw_text_det_res(dt_boxes, img_path):
LDOUBLEV's avatar
LDOUBLEV committed
186
187
188
189
    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
190
    return src_im
LDOUBLEV's avatar
LDOUBLEV committed
191
192


LDOUBLEV's avatar
LDOUBLEV committed
193
194
def resize_img(img, input_size=600):
    """
LDOUBLEV's avatar
LDOUBLEV committed
195
    resize img and limit the longest side of the image to input_size
LDOUBLEV's avatar
LDOUBLEV committed
196
197
198
199
200
    """
    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
201
202
    img = cv2.resize(img, None, None, fx=im_scale, fy=im_scale)
    return img
LDOUBLEV's avatar
LDOUBLEV committed
203
204


WenmuZhou's avatar
WenmuZhou committed
205
206
207
208
209
210
def draw_ocr(image,
             boxes,
             txts=None,
             scores=None,
             drop_score=0.5,
             font_path="./doc/simfang.ttf"):
211
212
213
    """
    Visualize the results of OCR detection and recognition
    args:
LDOUBLEV's avatar
LDOUBLEV committed
214
        image(Image|array): RGB image
215
216
217
218
        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
219
        font_path: the path of font which is used to draw text
220
221
222
    return(array):
        the visualized img
    """
LDOUBLEV's avatar
LDOUBLEV committed
223
224
    if scores is None:
        scores = [1] * len(boxes)
WenmuZhou's avatar
WenmuZhou committed
225
226
227
228
    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
229
            continue
WenmuZhou's avatar
WenmuZhou committed
230
        box = np.reshape(np.array(boxes[i]), [-1, 1, 2]).astype(np.int64)
LDOUBLEV's avatar
LDOUBLEV committed
231
        image = cv2.polylines(np.array(image), [box], True, (255, 0, 0), 2)
WenmuZhou's avatar
WenmuZhou committed
232
    if txts is not None:
LDOUBLEV's avatar
LDOUBLEV committed
233
        img = np.array(resize_img(image, input_size=600))
234
        txt_img = text_visual(
WenmuZhou's avatar
WenmuZhou committed
235
236
237
238
239
240
            txts,
            scores,
            img_h=img.shape[0],
            img_w=600,
            threshold=drop_score,
            font_path=font_path)
241
        img = np.concatenate([np.array(img), np.array(txt_img)], axis=1)
LDOUBLEV's avatar
LDOUBLEV committed
242
243
        return img
    return image
244
245


WenmuZhou's avatar
WenmuZhou committed
246
247
248
249
250
251
def draw_ocr_box_txt(image,
                     boxes,
                     txts,
                     scores=None,
                     drop_score=0.5,
                     font_path="./doc/simfang.ttf"):
252
253
254
    h, w = image.height, image.width
    img_left = image.copy()
    img_right = Image.new('RGB', (w, h), (255, 255, 255))
255
256

    import random
LDOUBLEV's avatar
LDOUBLEV committed
257

258
259
260
    random.seed(0)
    draw_left = ImageDraw.Draw(img_left)
    draw_right = ImageDraw.Draw(img_right)
WenmuZhou's avatar
WenmuZhou committed
261
262
263
    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
264
265
        color = (random.randint(0, 255), random.randint(0, 255),
                 random.randint(0, 255))
266
        draw_left.polygon(box, fill=color)
tink2123's avatar
tink2123 committed
267
268
269
270
271
272
273
274
275
276
        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)
277
278
        if box_height > 2 * box_width:
            font_size = max(int(box_width * 0.9), 10)
WenmuZhou's avatar
WenmuZhou committed
279
            font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
280
281
282
            cur_y = box[0][1]
            for c in txt:
                char_size = font.getsize(c)
tink2123's avatar
tink2123 committed
283
284
                draw_right.text(
                    (box[0][0] + 3, cur_y), c, fill=(0, 0, 0), font=font)
285
286
287
                cur_y += char_size[1]
        else:
            font_size = max(int(box_height * 0.8), 10)
WenmuZhou's avatar
WenmuZhou committed
288
            font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
tink2123's avatar
tink2123 committed
289
290
            draw_right.text(
                [box[0][0], box[0][1]], txt, fill=(0, 0, 0), font=font)
291
292
293
294
    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))
295
296
297
    return np.array(img_show)


298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
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
322
323
324
325
326
327
def text_visual(texts,
                scores,
                img_h=400,
                img_w=600,
                threshold=0.,
                font_path="./doc/simfang.ttf"):
328
329
330
331
332
333
334
    """
    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
335
        font_path: the path of font which is used to draw text
336
337
338
339
340
341
342
343
344
    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
345
346
        blank_img = Image.fromarray(blank_img).convert("RGB")
        draw_txt = ImageDraw.Draw(blank_img)
347
        return blank_img, draw_txt
LDOUBLEV's avatar
LDOUBLEV committed
348

349
350
351
352
    blank_img, draw_txt = create_blank_img()

    font_size = 20
    txt_color = (0, 0, 0)
WenmuZhou's avatar
WenmuZhou committed
353
    font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
354
355
356

    gap = font_size + 5
    txt_img_list = []
LDOUBLEV's avatar
LDOUBLEV committed
357
    count, index = 1, 0
358
359
    for idx, txt in enumerate(texts):
        index += 1
LDOUBLEV's avatar
LDOUBLEV committed
360
        if scores[idx] < threshold or math.isnan(scores[idx]):
361
362
363
364
365
366
367
368
369
370
371
            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
372
            draw_txt.text((0, gap * count), new_txt, txt_color, font=font)
373
374
375
376
377
            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
378
            count += 1
379
380
381
        if first_line:
            new_txt = str(index) + ': ' + txt + '   ' + '%.3f' % (scores[idx])
        else:
LDOUBLEV's avatar
LDOUBLEV committed
382
            new_txt = "  " + txt + "  " + '%.3f' % (scores[idx])
LDOUBLEV's avatar
LDOUBLEV committed
383
        draw_txt.text((0, gap * count), new_txt, txt_color, font=font)
384
        # whether add new blank img or not
LDOUBLEV's avatar
LDOUBLEV committed
385
        if count >= img_h // gap - 1 and idx + 1 < len(texts):
386
387
388
            txt_img_list.append(np.array(blank_img))
            blank_img, draw_txt = create_blank_img()
            count = 0
LDOUBLEV's avatar
LDOUBLEV committed
389
        count += 1
390
391
392
393
394
395
    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
396
397


dyning's avatar
dyning committed
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
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
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
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
434
    new_img = draw_ocr(image, boxes, txts, scores)
LDOUBLEV's avatar
LDOUBLEV committed
435

MissPenguin's avatar
MissPenguin committed
436
    cv2.imwrite(img_name, new_img)