utility.py 20.6 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
import time
from ppocr.utils.logging import get_logger
WenmuZhou's avatar
WenmuZhou committed
26

LDOUBLEV's avatar
LDOUBLEV committed
27
logger = get_logger()
LDOUBLEV's avatar
LDOUBLEV committed
28
29


30
31
def str2bool(v):
    return v.lower() in ("true", "t", "1")
LDOUBLEV's avatar
LDOUBLEV committed
32
33


WenmuZhou's avatar
WenmuZhou committed
34
def init_args():
LDOUBLEV's avatar
LDOUBLEV committed
35
    parser = argparse.ArgumentParser()
WenmuZhou's avatar
WenmuZhou committed
36
    # params for prediction engine
LDOUBLEV's avatar
LDOUBLEV committed
37
38
39
    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
40
    parser.add_argument("--min_subgraph_size", type=int, default=3)
LDOUBLEV's avatar
LDOUBLEV committed
41
    parser.add_argument("--precision", type=str, default="fp32")
42
    parser.add_argument("--gpu_mem", type=int, default=500)
LDOUBLEV's avatar
LDOUBLEV committed
43

WenmuZhou's avatar
WenmuZhou committed
44
    # params for text detector
LDOUBLEV's avatar
LDOUBLEV committed
45
46
47
    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
48
49
    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
50

WenmuZhou's avatar
WenmuZhou committed
51
    # DB parmas
LDOUBLEV's avatar
LDOUBLEV committed
52
53
    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
54
    parser.add_argument("--det_db_unclip_ratio", type=float, default=1.6)
LDOUBLEV's avatar
LDOUBLEV committed
55
    parser.add_argument("--max_batch_size", type=int, default=10)
LDOUBLEV's avatar
LDOUBLEV committed
56
    parser.add_argument("--use_dilation", type=bool, default=False)
littletomatodonkey's avatar
littletomatodonkey committed
57
    parser.add_argument("--det_db_score_mode", type=str, default="fast")
WenmuZhou's avatar
WenmuZhou committed
58
    # EAST parmas
LDOUBLEV's avatar
LDOUBLEV committed
59
60
61
62
    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
63
    # SAST parmas
licx's avatar
licx committed
64
65
    parser.add_argument("--det_sast_score_thresh", type=float, default=0.5)
    parser.add_argument("--det_sast_nms_thresh", type=float, default=0.2)
66
    parser.add_argument("--det_sast_polygon", type=bool, default=False)
licx's avatar
licx committed
67

WenmuZhou's avatar
WenmuZhou committed
68
    # params for text recognizer
LDOUBLEV's avatar
LDOUBLEV committed
69
70
    parser.add_argument("--rec_algorithm", type=str, default='CRNN')
    parser.add_argument("--rec_model_dir", type=str)
tink2123's avatar
fix bug  
tink2123 committed
71
72
    parser.add_argument("--rec_image_shape", type=str, default="3, 32, 320")
    parser.add_argument("--rec_char_type", type=str, default='ch')
73
    parser.add_argument("--rec_batch_num", type=int, default=6)
tink2123's avatar
fix bug  
tink2123 committed
74
    parser.add_argument("--max_text_length", type=int, default=25)
LDOUBLEV's avatar
LDOUBLEV committed
75
76
77
78
    parser.add_argument(
        "--rec_char_dict_path",
        type=str,
        default="./ppocr/utils/ppocr_keys_v1.txt")
WenmuZhou's avatar
WenmuZhou committed
79
80
    parser.add_argument("--use_space_char", type=str2bool, default=True)
    parser.add_argument(
tink2123's avatar
tink2123 committed
81
        "--vis_font_path", type=str, default="./doc/fonts/simfang.ttf")
WenmuZhou's avatar
WenmuZhou committed
82
    parser.add_argument("--drop_score", type=float, default=0.5)
WenmuZhou's avatar
WenmuZhou committed
83

Jethong's avatar
Jethong committed
84
85
86
87
88
89
90
91
92
    # 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
93
        "--e2e_char_dict_path", type=str, default="./ppocr/utils/ic15_dict.txt")
Jethong's avatar
Jethong committed
94
    parser.add_argument("--e2e_pgnet_valid_set", type=str, default='totaltext')
Jethong's avatar
Jethong committed
95
    parser.add_argument("--e2e_pgnet_polygon", type=bool, default=True)
Jethong's avatar
Jethong committed
96
    parser.add_argument("--e2e_pgnet_mode", type=str, default='fast')
Jethong's avatar
Jethong committed
97

WenmuZhou's avatar
WenmuZhou committed
98
99
100
101
102
    # 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'])
103
    parser.add_argument("--cls_batch_num", type=int, default=6)
WenmuZhou's avatar
WenmuZhou committed
104
105
106
    parser.add_argument("--cls_thresh", type=float, default=0.9)

    parser.add_argument("--enable_mkldnn", type=str2bool, default=False)
LDOUBLEV's avatar
LDOUBLEV committed
107
    parser.add_argument("--cpu_threads", type=int, default=10)
WenmuZhou's avatar
WenmuZhou committed
108
    parser.add_argument("--use_pdserving", type=str2bool, default=False)
LDOUBLEV's avatar
LDOUBLEV committed
109
    parser.add_argument("--warmup", type=str2bool, default=True)
WenmuZhou's avatar
WenmuZhou committed
110

LDOUBLEV's avatar
LDOUBLEV committed
111
    # multi-process
littletomatodonkey's avatar
littletomatodonkey committed
112
    parser.add_argument("--use_mp", type=str2bool, default=False)
113
114
    parser.add_argument("--total_process_num", type=int, default=1)
    parser.add_argument("--process_id", type=int, default=0)
WenmuZhou's avatar
WenmuZhou committed
115

LDOUBLEV's avatar
LDOUBLEV committed
116
117
    parser.add_argument("--benchmark", type=bool, default=False)
    parser.add_argument("--save_log_path", type=str, default="./log_output/")
Double_V's avatar
Double_V committed
118

WenmuZhou's avatar
WenmuZhou committed
119
    parser.add_argument("--show_log", type=str2bool, default=True)
WenmuZhou's avatar
WenmuZhou committed
120
    return parser
WenmuZhou's avatar
WenmuZhou committed
121

122

123
def parse_args():
WenmuZhou's avatar
WenmuZhou committed
124
    parser = init_args()
LDOUBLEV's avatar
LDOUBLEV committed
125
126
127
    return parser.parse_args()


WenmuZhou's avatar
WenmuZhou committed
128
129
130
131
132
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
133
    elif mode == 'rec':
WenmuZhou's avatar
WenmuZhou committed
134
        model_dir = args.rec_model_dir
WenmuZhou's avatar
WenmuZhou committed
135
136
    elif mode == 'table':
        model_dir = args.table_model_dir
Jethong's avatar
Jethong committed
137
138
    else:
        model_dir = args.e2e_model_dir
WenmuZhou's avatar
WenmuZhou committed
139
140
141
142

    if model_dir is None:
        logger.info("not find {} model file path {}".format(mode, model_dir))
        sys.exit(0)
WenmuZhou's avatar
WenmuZhou committed
143
144
    model_file_path = model_dir + "/inference.pdmodel"
    params_file_path = model_dir + "/inference.pdiparams"
WenmuZhou's avatar
WenmuZhou committed
145
    if not os.path.exists(model_file_path):
LDOUBLEV's avatar
LDOUBLEV committed
146
        raise ValueError("not find model file path {}".format(model_file_path))
WenmuZhou's avatar
WenmuZhou committed
147
    if not os.path.exists(params_file_path):
LDOUBLEV's avatar
LDOUBLEV committed
148
149
        raise ValueError("not find params file path {}".format(
            params_file_path))
WenmuZhou's avatar
WenmuZhou committed
150

WenmuZhou's avatar
WenmuZhou committed
151
    config = inference.Config(model_file_path, params_file_path)
WenmuZhou's avatar
WenmuZhou committed
152

LDOUBLEV's avatar
LDOUBLEV committed
153
154
155
156
157
158
159
160
161
162
    if hasattr(args, 'precision'):
        if args.precision == "fp16" and args.use_tensorrt:
            precision = inference.PrecisionType.Half
        elif args.precision == "int8":
            precision = inference.PrecisionType.Int8
        else:
            precision = inference.PrecisionType.Float32
    else:
        precision = inference.PrecisionType.Float32

WenmuZhou's avatar
WenmuZhou committed
163
164
    if args.use_gpu:
        config.enable_use_gpu(args.gpu_mem, 0)
LDOUBLEV's avatar
LDOUBLEV committed
165
166
        if args.use_tensorrt:
            config.enable_tensorrt_engine(
LDOUBLEV's avatar
LDOUBLEV committed
167
168
                precision_mode=inference.PrecisionType.Float32,
                max_batch_size=args.max_batch_size,
LDOUBLEV's avatar
LDOUBLEV committed
169
170
                min_subgraph_size=args.min_subgraph_size)
            # skip the minmum trt subgraph
LDOUBLEV's avatar
LDOUBLEV committed
171
        if mode == "det":
LDOUBLEV's avatar
LDOUBLEV committed
172
173
174
175
            min_input_shape = {
                "x": [1, 3, 50, 50],
                "conv2d_92.tmp_0": [1, 96, 20, 20],
                "conv2d_91.tmp_0": [1, 96, 10, 10],
LDOUBLEV's avatar
LDOUBLEV committed
176
                "conv2d_59.tmp_0": [1, 96, 20, 20],
LDOUBLEV's avatar
LDOUBLEV committed
177
178
                "nearest_interp_v2_1.tmp_0": [1, 96, 10, 10],
                "nearest_interp_v2_2.tmp_0": [1, 96, 20, 20],
LDOUBLEV's avatar
LDOUBLEV committed
179
                "conv2d_124.tmp_0": [1, 96, 20, 20],
LDOUBLEV's avatar
LDOUBLEV committed
180
181
182
183
184
185
186
187
188
189
                "nearest_interp_v2_3.tmp_0": [1, 24, 20, 20],
                "nearest_interp_v2_4.tmp_0": [1, 24, 20, 20],
                "nearest_interp_v2_5.tmp_0": [1, 24, 20, 20],
                "elementwise_add_7": [1, 56, 2, 2],
                "nearest_interp_v2_0.tmp_0": [1, 96, 2, 2]
            }
            max_input_shape = {
                "x": [1, 3, 2000, 2000],
                "conv2d_92.tmp_0": [1, 96, 400, 400],
                "conv2d_91.tmp_0": [1, 96, 200, 200],
LDOUBLEV's avatar
LDOUBLEV committed
190
                "conv2d_59.tmp_0": [1, 96, 400, 400],
LDOUBLEV's avatar
LDOUBLEV committed
191
                "nearest_interp_v2_1.tmp_0": [1, 96, 200, 200],
LDOUBLEV's avatar
LDOUBLEV committed
192
                "conv2d_124.tmp_0": [1, 256, 400, 400],
LDOUBLEV's avatar
LDOUBLEV committed
193
194
195
196
197
198
199
200
201
202
203
                "nearest_interp_v2_2.tmp_0": [1, 96, 400, 400],
                "nearest_interp_v2_3.tmp_0": [1, 24, 400, 400],
                "nearest_interp_v2_4.tmp_0": [1, 24, 400, 400],
                "nearest_interp_v2_5.tmp_0": [1, 24, 400, 400],
                "elementwise_add_7": [1, 56, 400, 400],
                "nearest_interp_v2_0.tmp_0": [1, 96, 400, 400]
            }
            opt_input_shape = {
                "x": [1, 3, 640, 640],
                "conv2d_92.tmp_0": [1, 96, 160, 160],
                "conv2d_91.tmp_0": [1, 96, 80, 80],
LDOUBLEV's avatar
LDOUBLEV committed
204
                "conv2d_59.tmp_0": [1, 96, 160, 160],
LDOUBLEV's avatar
LDOUBLEV committed
205
206
                "nearest_interp_v2_1.tmp_0": [1, 96, 80, 80],
                "nearest_interp_v2_2.tmp_0": [1, 96, 160, 160],
LDOUBLEV's avatar
LDOUBLEV committed
207
                "conv2d_124.tmp_0": [1, 256, 160, 160],
LDOUBLEV's avatar
LDOUBLEV committed
208
209
210
211
212
213
214
215
216
217
218
219
220
221
                "nearest_interp_v2_3.tmp_0": [1, 24, 160, 160],
                "nearest_interp_v2_4.tmp_0": [1, 24, 160, 160],
                "nearest_interp_v2_5.tmp_0": [1, 24, 160, 160],
                "elementwise_add_7": [1, 56, 40, 40],
                "nearest_interp_v2_0.tmp_0": [1, 96, 40, 40]
            }
        elif mode == "rec":
            min_input_shape = {"x": [args.rec_batch_num, 3, 32, 10]}
            max_input_shape = {"x": [args.rec_batch_num, 3, 32, 2000]}
            opt_input_shape = {"x": [args.rec_batch_num, 3, 32, 320]}
        elif mode == "cls":
            min_input_shape = {"x": [args.rec_batch_num, 3, 48, 10]}
            max_input_shape = {"x": [args.rec_batch_num, 3, 48, 2000]}
            opt_input_shape = {"x": [args.rec_batch_num, 3, 48, 320]}
LDOUBLEV's avatar
LDOUBLEV committed
222
223
224
225
        else:
            min_input_shape = {"x": [1, 3, 10, 10]}
            max_input_shape = {"x": [1, 3, 1000, 1000]}
            opt_input_shape = {"x": [1, 3, 500, 500]}
LDOUBLEV's avatar
LDOUBLEV committed
226
227
228
        config.set_trt_dynamic_shape_info(min_input_shape, max_input_shape,
                                          opt_input_shape)

WenmuZhou's avatar
WenmuZhou committed
229
230
    else:
        config.disable_gpu()
LDOUBLEV's avatar
LDOUBLEV committed
231
232
233
        if hasattr(args, "cpu_threads"):
            config.set_cpu_math_library_num_threads(args.cpu_threads)
        else:
WenmuZhou's avatar
WenmuZhou committed
234
            # default cpu threads as 10
LDOUBLEV's avatar
LDOUBLEV committed
235
            config.set_cpu_math_library_num_threads(10)
WenmuZhou's avatar
WenmuZhou committed
236
237
238
239
240
        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
241
242
    # enable memory optim
    config.enable_memory_optim()
LDOUBLEV's avatar
LDOUBLEV committed
243
    #config.disable_glog_info()
WenmuZhou's avatar
WenmuZhou committed
244

WenmuZhou's avatar
WenmuZhou committed
245
    config.delete_pass("conv_transpose_eltwiseadd_bn_fuse_pass")
WenmuZhou's avatar
WenmuZhou committed
246
    if mode == 'table':
WenmuZhou's avatar
WenmuZhou committed
247
        config.delete_pass("fc_fuse_pass")  # not supported for table
WenmuZhou's avatar
WenmuZhou committed
248
    config.switch_use_feed_fetch_ops(False)
WenmuZhou's avatar
WenmuZhou committed
249
    config.switch_ir_optim(True)
250

WenmuZhou's avatar
WenmuZhou committed
251
252
    # create predictor
    predictor = inference.create_predictor(config)
WenmuZhou's avatar
WenmuZhou committed
253
254
    input_names = predictor.get_input_names()
    for name in input_names:
WenmuZhou's avatar
WenmuZhou committed
255
        input_tensor = predictor.get_input_handle(name)
WenmuZhou's avatar
WenmuZhou committed
256
257
258
    output_names = predictor.get_output_names()
    output_tensors = []
    for output_name in output_names:
WenmuZhou's avatar
WenmuZhou committed
259
        output_tensor = predictor.get_output_handle(output_name)
WenmuZhou's avatar
WenmuZhou committed
260
        output_tensors.append(output_tensor)
LDOUBLEV's avatar
LDOUBLEV committed
261
    return predictor, input_tensor, output_tensors, config
WenmuZhou's avatar
WenmuZhou committed
262
263


Jethong's avatar
Jethong committed
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
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
280
def draw_text_det_res(dt_boxes, img_path):
LDOUBLEV's avatar
LDOUBLEV committed
281
282
283
284
    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
285
    return src_im
LDOUBLEV's avatar
LDOUBLEV committed
286
287


LDOUBLEV's avatar
LDOUBLEV committed
288
289
def resize_img(img, input_size=600):
    """
LDOUBLEV's avatar
LDOUBLEV committed
290
    resize img and limit the longest side of the image to input_size
LDOUBLEV's avatar
LDOUBLEV committed
291
292
293
294
295
    """
    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
296
297
    img = cv2.resize(img, None, None, fx=im_scale, fy=im_scale)
    return img
LDOUBLEV's avatar
LDOUBLEV committed
298
299


WenmuZhou's avatar
WenmuZhou committed
300
301
302
303
304
def draw_ocr(image,
             boxes,
             txts=None,
             scores=None,
             drop_score=0.5,
LDOUBLEV's avatar
LDOUBLEV committed
305
             font_path="./doc/fonts/simfang.ttf"):
306
307
308
    """
    Visualize the results of OCR detection and recognition
    args:
LDOUBLEV's avatar
LDOUBLEV committed
309
        image(Image|array): RGB image
310
311
312
313
        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
314
        font_path: the path of font which is used to draw text
315
316
317
    return(array):
        the visualized img
    """
LDOUBLEV's avatar
LDOUBLEV committed
318
319
    if scores is None:
        scores = [1] * len(boxes)
WenmuZhou's avatar
WenmuZhou committed
320
321
322
323
    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
324
            continue
WenmuZhou's avatar
WenmuZhou committed
325
        box = np.reshape(np.array(boxes[i]), [-1, 1, 2]).astype(np.int64)
LDOUBLEV's avatar
LDOUBLEV committed
326
        image = cv2.polylines(np.array(image), [box], True, (255, 0, 0), 2)
WenmuZhou's avatar
WenmuZhou committed
327
    if txts is not None:
LDOUBLEV's avatar
LDOUBLEV committed
328
        img = np.array(resize_img(image, input_size=600))
329
        txt_img = text_visual(
WenmuZhou's avatar
WenmuZhou committed
330
331
332
333
334
335
            txts,
            scores,
            img_h=img.shape[0],
            img_w=600,
            threshold=drop_score,
            font_path=font_path)
336
        img = np.concatenate([np.array(img), np.array(txt_img)], axis=1)
LDOUBLEV's avatar
LDOUBLEV committed
337
338
        return img
    return image
339
340


WenmuZhou's avatar
WenmuZhou committed
341
342
343
344
345
346
def draw_ocr_box_txt(image,
                     boxes,
                     txts,
                     scores=None,
                     drop_score=0.5,
                     font_path="./doc/simfang.ttf"):
347
348
349
    h, w = image.height, image.width
    img_left = image.copy()
    img_right = Image.new('RGB', (w, h), (255, 255, 255))
350
351

    import random
LDOUBLEV's avatar
LDOUBLEV committed
352

353
354
355
    random.seed(0)
    draw_left = ImageDraw.Draw(img_left)
    draw_right = ImageDraw.Draw(img_right)
WenmuZhou's avatar
WenmuZhou committed
356
357
358
    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
359
360
        color = (random.randint(0, 255), random.randint(0, 255),
                 random.randint(0, 255))
361
        draw_left.polygon(box, fill=color)
tink2123's avatar
tink2123 committed
362
363
364
365
366
367
368
369
370
371
        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)
372
373
        if box_height > 2 * box_width:
            font_size = max(int(box_width * 0.9), 10)
WenmuZhou's avatar
WenmuZhou committed
374
            font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
375
376
377
            cur_y = box[0][1]
            for c in txt:
                char_size = font.getsize(c)
tink2123's avatar
tink2123 committed
378
379
                draw_right.text(
                    (box[0][0] + 3, cur_y), c, fill=(0, 0, 0), font=font)
380
381
382
                cur_y += char_size[1]
        else:
            font_size = max(int(box_height * 0.8), 10)
WenmuZhou's avatar
WenmuZhou committed
383
            font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
tink2123's avatar
tink2123 committed
384
385
            draw_right.text(
                [box[0][0], box[0][1]], txt, fill=(0, 0, 0), font=font)
386
387
388
389
    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))
390
391
392
    return np.array(img_show)


393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
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
417
418
419
420
421
422
def text_visual(texts,
                scores,
                img_h=400,
                img_w=600,
                threshold=0.,
                font_path="./doc/simfang.ttf"):
423
424
425
426
427
428
429
    """
    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
430
        font_path: the path of font which is used to draw text
431
432
433
434
435
436
437
438
439
    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
440
441
        blank_img = Image.fromarray(blank_img).convert("RGB")
        draw_txt = ImageDraw.Draw(blank_img)
442
        return blank_img, draw_txt
LDOUBLEV's avatar
LDOUBLEV committed
443

444
445
446
447
    blank_img, draw_txt = create_blank_img()

    font_size = 20
    txt_color = (0, 0, 0)
WenmuZhou's avatar
WenmuZhou committed
448
    font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
449
450
451

    gap = font_size + 5
    txt_img_list = []
LDOUBLEV's avatar
LDOUBLEV committed
452
    count, index = 1, 0
453
454
    for idx, txt in enumerate(texts):
        index += 1
LDOUBLEV's avatar
LDOUBLEV committed
455
        if scores[idx] < threshold or math.isnan(scores[idx]):
456
457
458
459
460
461
462
463
464
465
466
            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
467
            draw_txt.text((0, gap * count), new_txt, txt_color, font=font)
468
469
470
471
472
            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
473
            count += 1
474
475
476
        if first_line:
            new_txt = str(index) + ': ' + txt + '   ' + '%.3f' % (scores[idx])
        else:
LDOUBLEV's avatar
LDOUBLEV committed
477
            new_txt = "  " + txt + "  " + '%.3f' % (scores[idx])
LDOUBLEV's avatar
LDOUBLEV committed
478
        draw_txt.text((0, gap * count), new_txt, txt_color, font=font)
479
        # whether add new blank img or not
LDOUBLEV's avatar
LDOUBLEV committed
480
        if count >= img_h // gap - 1 and idx + 1 < len(texts):
481
482
483
            txt_img_list.append(np.array(blank_img))
            blank_img, draw_txt = create_blank_img()
            count = 0
LDOUBLEV's avatar
LDOUBLEV committed
484
        count += 1
485
486
487
488
489
490
    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
491
492


dyning's avatar
dyning committed
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
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


WenmuZhou's avatar
WenmuZhou committed
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
def get_rotate_crop_image(img, points):
    '''
    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
    '''
    assert len(points) == 4, "shape of points must be 4*2"
    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],
                          [img_crop_width, img_crop_height],
                          [0, img_crop_height]])
    M = cv2.getPerspectiveTransform(points, pts_std)
    dst_img = cv2.warpPerspective(
        img,
        M, (img_crop_width, img_crop_height),
        borderMode=cv2.BORDER_REPLICATE,
        flags=cv2.INTER_CUBIC)
    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


LDOUBLEV's avatar
LDOUBLEV committed
547
if __name__ == '__main__':
LDOUBLEV's avatar
LDOUBLEV committed
548
    pass