utility.py 14.4 KB
Newer Older
L
LDOUBLEV 已提交
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
W
WenmuZhou 已提交
16
import os
W
WenmuZhou 已提交
17
import sys
L
LDOUBLEV 已提交
18 19
import cv2
import numpy as np
L
LDOUBLEV 已提交
20 21
import json
from PIL import Image, ImageDraw, ImageFont
22
import math
W
WenmuZhou 已提交
23
from paddle import inference
L
LDOUBLEV 已提交
24 25


26 27
def str2bool(v):
    return v.lower() in ("true", "t", "1")
L
LDOUBLEV 已提交
28 29


30
inference_args_list = [
31
    # name     type      defalue
32
    # params for prediction engine
33 34 35 36 37 38 39 40 41 42
    ['use_gpu', str2bool, True],
    ['use_tensorrt', str2bool, False],
    ['use_fp16', str2bool, False],
    ['use_pdserving', str2bool, False],
    ['use_mp', str2bool, False],
    ['enable_mkldnn', str2bool, False],
    ['ir_optim', str2bool, True],
    ['total_process_num', int, 1],
    ['process_id', int, 0],
    ['gpu_mem', int, 500],
W
WenmuZhou 已提交
43
    # params for text detector
44 45 46 47 48
    ['image_dir', str, None],
    ['det_algorithm', str, 'DB'],
    ['det_model_dir', str, None],
    ['det_limit_side_len', float, 960],
    ['det_limit_type', str, 'max'],
W
WenmuZhou 已提交
49
    # DB parmas
50 51 52 53 54 55
    ['det_db_thresh', float, 0.3],
    ['det_db_box_thresh', float, 0.5],
    ['det_db_unclip_ratio', float, 1.6],
    ['max_batch_size', int, 10],
    ['use_dilation', str2bool, False],
    ['det_db_score_mode', str, 'fast'],
W
WenmuZhou 已提交
56
    # EAST parmas
57 58 59
    ['det_east_score_thresh', float, 0.8],
    ['det_east_cover_thresh', float, 0.1],
    ['det_east_nms_thresh', float, 0.2],
W
WenmuZhou 已提交
60
    # SAST parmas
61 62 63
    ['det_sast_score_thresh', float, 0.5],
    ['det_sast_nms_thresh', float, 0.2],
    ['det_sast_polygon', str2bool, False],
W
WenmuZhou 已提交
64
    # params for text recognizer
65 66 67 68 69 70 71 72 73 74
    ['rec_algorithm', str, 'CRNN'],
    ['rec_model_dir', str, None],
    ['rec_image_shape', str, '3, 32, 320'],
    ['rec_char_type', str, "ch"],
    ['rec_batch_num', int, 6],
    ['max_text_length', int, 25],
    ['rec_char_dict_path', str, './ppocr/utils/ppocr_keys_v1.txt'],
    ['use_space_char', str2bool, True],
    ['vis_font_path', str, './doc/fonts/simfang.ttf'],
    ['drop_score', float, 0.5],
J
Jethong 已提交
75
    # params for e2e
76 77 78 79
    ['e2e_algorithm', str, 'PGNet'],
    ['e2e_model_dir', str, None],
    ['e2e_limit_side_len', float, 768],
    ['e2e_limit_type', str, 'max'],
J
Jethong 已提交
80
    # PGNet parmas
81 82 83 84 85
    ['e2e_pgnet_score_thresh', float, 0.5],
    ['e2e_char_dict_path', str, './ppocr/utils/ic15_dict.txt'],
    ['e2e_pgnet_valid_set', str, 'totaltext'],
    ['e2e_pgnet_polygon', str2bool, True],
    ['e2e_pgnet_mode', str, 'fast'],
W
WenmuZhou 已提交
86
    # params for text classifier
87 88 89 90 91 92
    ['use_angle_cls', str2bool, False],
    ['cls_model_dir', str, None],
    ['cls_image_shape', str, '3, 48, 192'],
    ['label_list', list, ['0', '180']],
    ['cls_batch_num', int, 6],
    ['cls_thresh', float, 0.9],
93
]
W
WenmuZhou 已提交
94

95

96 97 98
def parse_args():
    parser = argparse.ArgumentParser()
    for item in inference_args_list:
99
        parser.add_argument('--' + item[0], type=item[1], default=item[2])
L
LDOUBLEV 已提交
100 101 102
    return parser.parse_args()


W
WenmuZhou 已提交
103 104 105 106 107
def create_predictor(args, mode, logger):
    if mode == "det":
        model_dir = args.det_model_dir
    elif mode == 'cls':
        model_dir = args.cls_model_dir
J
Jethong 已提交
108
    elif mode == 'rec':
W
WenmuZhou 已提交
109
        model_dir = args.rec_model_dir
J
Jethong 已提交
110 111
    else:
        model_dir = args.e2e_model_dir
W
WenmuZhou 已提交
112 113 114 115

    if model_dir is None:
        logger.info("not find {} model file path {}".format(mode, model_dir))
        sys.exit(0)
文幕地方's avatar
文幕地方 已提交
116 117
    model_file_path = model_dir + "/inference.pdmodel"
    params_file_path = model_dir + "/inference.pdiparams"
W
WenmuZhou 已提交
118 119 120 121 122 123 124
    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)

W
WenmuZhou 已提交
125
    config = inference.Config(model_file_path, params_file_path)
W
WenmuZhou 已提交
126 127 128

    if args.use_gpu:
        config.enable_use_gpu(args.gpu_mem, 0)
L
LDOUBLEV 已提交
129 130
        if args.use_tensorrt:
            config.enable_tensorrt_engine(
W
WenmuZhou 已提交
131 132
                precision_mode=inference.PrecisionType.Half
                if args.use_fp16 else inference.PrecisionType.Float32,
L
LDOUBLEV 已提交
133
                max_batch_size=args.max_batch_size)
W
WenmuZhou 已提交
134 135 136 137 138 139 140
    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()
L
LDOUBLEV 已提交
141
            #  TODO LDOUBLEV: fix mkldnn bug when bach_size  > 1
142
            # config.set_mkldnn_op({'conv2d', 'depthwise_conv2d', 'pool2d', 'batch_norm'})
L
LDOUBLEV 已提交
143
            args.rec_batch_num = 1
W
WenmuZhou 已提交
144

L
LDOUBLEV 已提交
145 146
    # enable memory optim
    config.enable_memory_optim()
W
WenmuZhou 已提交
147 148
    config.disable_glog_info()

W
WenmuZhou 已提交
149 150
    config.delete_pass("conv_transpose_eltwiseadd_bn_fuse_pass")
    config.switch_use_feed_fetch_ops(False)
W
WenmuZhou 已提交
151

W
WenmuZhou 已提交
152 153
    # create predictor
    predictor = inference.create_predictor(config)
W
WenmuZhou 已提交
154 155
    input_names = predictor.get_input_names()
    for name in input_names:
W
WenmuZhou 已提交
156
        input_tensor = predictor.get_input_handle(name)
W
WenmuZhou 已提交
157 158 159
    output_names = predictor.get_output_names()
    output_tensors = []
    for output_name in output_names:
W
WenmuZhou 已提交
160
        output_tensor = predictor.get_output_handle(output_name)
W
WenmuZhou 已提交
161 162 163 164
        output_tensors.append(output_tensor)
    return predictor, input_tensor, output_tensors


J
Jethong 已提交
165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180
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


L
LDOUBLEV 已提交
181
def draw_text_det_res(dt_boxes, img_path):
L
LDOUBLEV 已提交
182 183 184 185
    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)
L
LDOUBLEV 已提交
186
    return src_im
L
LDOUBLEV 已提交
187 188


L
LDOUBLEV 已提交
189 190
def resize_img(img, input_size=600):
    """
L
LDOUBLEV 已提交
191
    resize img and limit the longest side of the image to input_size
L
LDOUBLEV 已提交
192 193 194 195 196
    """
    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)
W
WenmuZhou 已提交
197 198
    img = cv2.resize(img, None, None, fx=im_scale, fy=im_scale)
    return img
L
LDOUBLEV 已提交
199 200


W
WenmuZhou 已提交
201 202 203 204 205 206
def draw_ocr(image,
             boxes,
             txts=None,
             scores=None,
             drop_score=0.5,
             font_path="./doc/simfang.ttf"):
207 208 209
    """
    Visualize the results of OCR detection and recognition
    args:
L
LDOUBLEV 已提交
210
        image(Image|array): RGB image
211 212 213 214
        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
W
WenmuZhou 已提交
215
        font_path: the path of font which is used to draw text
216 217 218
    return(array):
        the visualized img
    """
L
LDOUBLEV 已提交
219 220
    if scores is None:
        scores = [1] * len(boxes)
W
WenmuZhou 已提交
221 222 223 224
    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])):
L
LDOUBLEV 已提交
225
            continue
W
WenmuZhou 已提交
226
        box = np.reshape(np.array(boxes[i]), [-1, 1, 2]).astype(np.int64)
L
LDOUBLEV 已提交
227
        image = cv2.polylines(np.array(image), [box], True, (255, 0, 0), 2)
W
WenmuZhou 已提交
228
    if txts is not None:
L
LDOUBLEV 已提交
229
        img = np.array(resize_img(image, input_size=600))
230
        txt_img = text_visual(
W
WenmuZhou 已提交
231 232 233 234 235 236
            txts,
            scores,
            img_h=img.shape[0],
            img_w=600,
            threshold=drop_score,
            font_path=font_path)
237
        img = np.concatenate([np.array(img), np.array(txt_img)], axis=1)
L
LDOUBLEV 已提交
238 239
        return img
    return image
240 241


W
WenmuZhou 已提交
242 243 244 245 246 247
def draw_ocr_box_txt(image,
                     boxes,
                     txts,
                     scores=None,
                     drop_score=0.5,
                     font_path="./doc/simfang.ttf"):
248 249 250
    h, w = image.height, image.width
    img_left = image.copy()
    img_right = Image.new('RGB', (w, h), (255, 255, 255))
251 252

    import random
L
LDOUBLEV 已提交
253

254 255 256
    random.seed(0)
    draw_left = ImageDraw.Draw(img_left)
    draw_right = ImageDraw.Draw(img_right)
W
WenmuZhou 已提交
257 258 259
    for idx, (box, txt) in enumerate(zip(boxes, txts)):
        if scores is not None and scores[idx] < drop_score:
            continue
T
tink2123 已提交
260 261
        color = (random.randint(0, 255), random.randint(0, 255),
                 random.randint(0, 255))
262
        draw_left.polygon(box, fill=color)
T
tink2123 已提交
263 264 265 266 267 268 269 270 271 272
        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)
273 274
        if box_height > 2 * box_width:
            font_size = max(int(box_width * 0.9), 10)
W
WenmuZhou 已提交
275
            font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
276 277 278
            cur_y = box[0][1]
            for c in txt:
                char_size = font.getsize(c)
T
tink2123 已提交
279 280
                draw_right.text(
                    (box[0][0] + 3, cur_y), c, fill=(0, 0, 0), font=font)
281 282 283
                cur_y += char_size[1]
        else:
            font_size = max(int(box_height * 0.8), 10)
W
WenmuZhou 已提交
284
            font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
T
tink2123 已提交
285 286
            draw_right.text(
                [box[0][0], box[0][1]], txt, fill=(0, 0, 0), font=font)
287 288 289 290
    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))
291 292 293
    return np.array(img_show)


294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317
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)


W
WenmuZhou 已提交
318 319 320 321 322 323
def text_visual(texts,
                scores,
                img_h=400,
                img_w=600,
                threshold=0.,
                font_path="./doc/simfang.ttf"):
324 325 326 327 328 329 330
    """
    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
W
WenmuZhou 已提交
331
        font_path: the path of font which is used to draw text
332 333 334 335 336 337 338 339 340
    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
L
LDOUBLEV 已提交
341 342
        blank_img = Image.fromarray(blank_img).convert("RGB")
        draw_txt = ImageDraw.Draw(blank_img)
343
        return blank_img, draw_txt
L
LDOUBLEV 已提交
344

345 346 347 348
    blank_img, draw_txt = create_blank_img()

    font_size = 20
    txt_color = (0, 0, 0)
W
WenmuZhou 已提交
349
    font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
350 351 352

    gap = font_size + 5
    txt_img_list = []
L
LDOUBLEV 已提交
353
    count, index = 1, 0
354 355
    for idx, txt in enumerate(texts):
        index += 1
L
LDOUBLEV 已提交
356
        if scores[idx] < threshold or math.isnan(scores[idx]):
357 358 359 360 361 362 363 364 365 366 367
            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
L
LDOUBLEV 已提交
368
            draw_txt.text((0, gap * count), new_txt, txt_color, font=font)
369 370 371 372 373
            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
L
LDOUBLEV 已提交
374
            count += 1
375 376 377
        if first_line:
            new_txt = str(index) + ': ' + txt + '   ' + '%.3f' % (scores[idx])
        else:
L
LDOUBLEV 已提交
378
            new_txt = "  " + txt + "  " + '%.3f' % (scores[idx])
L
LDOUBLEV 已提交
379
        draw_txt.text((0, gap * count), new_txt, txt_color, font=font)
380
        # whether add new blank img or not
L
LDOUBLEV 已提交
381
        if count >= img_h // gap - 1 and idx + 1 < len(texts):
382 383 384
            txt_img_list.append(np.array(blank_img))
            blank_img, draw_txt = create_blank_img()
            count = 0
L
LDOUBLEV 已提交
385
        count += 1
386 387 388 389 390 391
    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)
L
LDOUBLEV 已提交
392 393


D
dyning 已提交
394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412
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


L
LDOUBLEV 已提交
413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429
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))

W
WenmuZhou 已提交
430
    new_img = draw_ocr(image, boxes, txts, scores)
L
LDOUBLEV 已提交
431

M
MissPenguin 已提交
432
    cv2.imwrite(img_name, new_img)