predict_rec.py 15.4 KB
Newer Older
L
LDOUBLEV 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
L
LDOUBLEV 已提交
14 15
import os
import sys
T
Topdu 已提交
16
from PIL import Image
17
__dir__ = os.path.dirname(os.path.abspath(__file__))
L
LDOUBLEV 已提交
18
sys.path.append(__dir__)
19
sys.path.append(os.path.abspath(os.path.join(__dir__, '../..')))
L
LDOUBLEV 已提交
20

L
LDOUBLEV 已提交
21 22
os.environ["FLAGS_allocator_strategy"] = 'auto_growth'

L
LDOUBLEV 已提交
23 24 25 26
import cv2
import numpy as np
import math
import time
W
WenmuZhou 已提交
27
import traceback
T
tink2123 已提交
28
import paddle
29 30

import tools.infer.utility as utility
W
WenmuZhou 已提交
31 32
from ppocr.postprocess import build_post_process
from ppocr.utils.logging import get_logger
33
from ppocr.utils.utility import get_image_file_list, check_and_read_gif
L
LDOUBLEV 已提交
34

W
WenmuZhou 已提交
35 36
logger = get_logger()

L
LDOUBLEV 已提交
37 38 39

class TextRecognizer(object):
    def __init__(self, args):
40
        self.rec_image_shape = [int(v) for v in args.rec_image_shape.split(",")]
41
        self.rec_batch_num = args.rec_batch_num
T
tink2123 已提交
42
        self.rec_algorithm = args.rec_algorithm
W
WenmuZhou 已提交
43 44
        postprocess_params = {
            'name': 'CTCLabelDecode',
45
            "character_dict_path": args.rec_char_dict_path,
W
WenmuZhou 已提交
46
            "use_space_char": args.use_space_char
T
tink2123 已提交
47
        }
T
tink2123 已提交
48 49 50
        if self.rec_algorithm == "SRN":
            postprocess_params = {
                'name': 'SRNLabelDecode',
W
WenmuZhou 已提交
51 52 53 54 55 56
                "character_dict_path": args.rec_char_dict_path,
                "use_space_char": args.use_space_char
            }
        elif self.rec_algorithm == "RARE":
            postprocess_params = {
                'name': 'AttnLabelDecode',
T
tink2123 已提交
57 58 59
                "character_dict_path": args.rec_char_dict_path,
                "use_space_char": args.use_space_char
            }
T
Topdu 已提交
60 61 62 63 64 65
        elif self.rec_algorithm == 'NRTR':
            postprocess_params = {
                'name': 'NRTRLabelDecode',
                "character_dict_path": args.rec_char_dict_path,
                "use_space_char": args.use_space_char
            }
T
Topdu 已提交
66 67 68 69 70 71
        elif self.rec_algorithm == "SAR":
            postprocess_params = {
                'name': 'SARLabelDecode',
                "character_dict_path": args.rec_char_dict_path,
                "use_space_char": args.use_space_char
            }
W
WenmuZhou 已提交
72
        self.postprocess_op = build_post_process(postprocess_params)
L
LDOUBLEV 已提交
73
        self.predictor, self.input_tensor, self.output_tensors, self.config = \
W
WenmuZhou 已提交
74
            utility.create_predictor(args, 'rec', logger)
T
tink2123 已提交
75 76 77 78
        self.benchmark = args.benchmark
        if args.benchmark:
            import auto_log
            pid = os.getpid()
L
LDOUBLEV 已提交
79
            gpu_id = utility.get_infer_gpuid()
T
tink2123 已提交
80 81 82
            self.autolog = auto_log.AutoLogger(
                model_name="rec",
                model_precision=args.precision,
T
tink2123 已提交
83
                batch_size=args.rec_batch_num,
T
tink2123 已提交
84
                data_shape="dynamic",
85
                save_path=None,  #args.save_log_path,
T
tink2123 已提交
86 87 88
                inference_config=self.config,
                pids=pid,
                process_name=None,
L
LDOUBLEV 已提交
89
                gpu_ids=gpu_id if args.use_gpu else None,
T
tink2123 已提交
90 91 92
                time_keys=[
                    'preprocess_time', 'inference_time', 'postprocess_time'
                ],
93 94
                warmup=2,
                logger=logger)
L
LDOUBLEV 已提交
95

96
    def resize_norm_img(self, img, max_wh_ratio):
L
LDOUBLEV 已提交
97
        imgC, imgH, imgW = self.rec_image_shape
T
Topdu 已提交
98
        if self.rec_algorithm == 'NRTR':
T
Topdu 已提交
99 100 101 102 103 104 105 106 107
            img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
            # return padding_im
            image_pil = Image.fromarray(np.uint8(img))
            img = image_pil.resize([100, 32], Image.ANTIALIAS)
            img = np.array(img)
            norm_img = np.expand_dims(img, -1)
            norm_img = norm_img.transpose((2, 0, 1))
            return norm_img.astype(np.float32) / 128. - 1.

108
        assert imgC == img.shape[2]
T
tink2123 已提交
109
        imgW = int((32 * max_wh_ratio))
110
        h, w = img.shape[:2]
111 112 113 114 115
        ratio = w / float(h)
        if math.ceil(imgH * ratio) > imgW:
            resized_w = imgW
        else:
            resized_w = int(math.ceil(imgH * ratio))
T
tink2123 已提交
116
        resized_image = cv2.resize(img, (resized_w, imgH))
L
LDOUBLEV 已提交
117 118 119 120 121 122 123 124
        resized_image = resized_image.astype('float32')
        resized_image = resized_image.transpose((2, 0, 1)) / 255
        resized_image -= 0.5
        resized_image /= 0.5
        padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
        padding_im[:, :, 0:resized_w] = resized_image
        return padding_im

T
tink2123 已提交
125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196
    def resize_norm_img_srn(self, img, image_shape):
        imgC, imgH, imgW = image_shape

        img_black = np.zeros((imgH, imgW))
        im_hei = img.shape[0]
        im_wid = img.shape[1]

        if im_wid <= im_hei * 1:
            img_new = cv2.resize(img, (imgH * 1, imgH))
        elif im_wid <= im_hei * 2:
            img_new = cv2.resize(img, (imgH * 2, imgH))
        elif im_wid <= im_hei * 3:
            img_new = cv2.resize(img, (imgH * 3, imgH))
        else:
            img_new = cv2.resize(img, (imgW, imgH))

        img_np = np.asarray(img_new)
        img_np = cv2.cvtColor(img_np, cv2.COLOR_BGR2GRAY)
        img_black[:, 0:img_np.shape[1]] = img_np
        img_black = img_black[:, :, np.newaxis]

        row, col, c = img_black.shape
        c = 1

        return np.reshape(img_black, (c, row, col)).astype(np.float32)

    def srn_other_inputs(self, image_shape, num_heads, max_text_length):

        imgC, imgH, imgW = image_shape
        feature_dim = int((imgH / 8) * (imgW / 8))

        encoder_word_pos = np.array(range(0, feature_dim)).reshape(
            (feature_dim, 1)).astype('int64')
        gsrm_word_pos = np.array(range(0, max_text_length)).reshape(
            (max_text_length, 1)).astype('int64')

        gsrm_attn_bias_data = np.ones((1, max_text_length, max_text_length))
        gsrm_slf_attn_bias1 = np.triu(gsrm_attn_bias_data, 1).reshape(
            [-1, 1, max_text_length, max_text_length])
        gsrm_slf_attn_bias1 = np.tile(
            gsrm_slf_attn_bias1,
            [1, num_heads, 1, 1]).astype('float32') * [-1e9]

        gsrm_slf_attn_bias2 = np.tril(gsrm_attn_bias_data, -1).reshape(
            [-1, 1, max_text_length, max_text_length])
        gsrm_slf_attn_bias2 = np.tile(
            gsrm_slf_attn_bias2,
            [1, num_heads, 1, 1]).astype('float32') * [-1e9]

        encoder_word_pos = encoder_word_pos[np.newaxis, :]
        gsrm_word_pos = gsrm_word_pos[np.newaxis, :]

        return [
            encoder_word_pos, gsrm_word_pos, gsrm_slf_attn_bias1,
            gsrm_slf_attn_bias2
        ]

    def process_image_srn(self, img, image_shape, num_heads, max_text_length):
        norm_img = self.resize_norm_img_srn(img, image_shape)
        norm_img = norm_img[np.newaxis, :]

        [encoder_word_pos, gsrm_word_pos, gsrm_slf_attn_bias1, gsrm_slf_attn_bias2] = \
            self.srn_other_inputs(image_shape, num_heads, max_text_length)

        gsrm_slf_attn_bias1 = gsrm_slf_attn_bias1.astype(np.float32)
        gsrm_slf_attn_bias2 = gsrm_slf_attn_bias2.astype(np.float32)
        encoder_word_pos = encoder_word_pos.astype(np.int64)
        gsrm_word_pos = gsrm_word_pos.astype(np.int64)

        return (norm_img, encoder_word_pos, gsrm_word_pos, gsrm_slf_attn_bias1,
                gsrm_slf_attn_bias2)

T
Topdu 已提交
197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231
    def resize_norm_img_sar(self, img, image_shape,
                            width_downsample_ratio=0.25):
        imgC, imgH, imgW_min, imgW_max = image_shape
        h = img.shape[0]
        w = img.shape[1]
        valid_ratio = 1.0
        # make sure new_width is an integral multiple of width_divisor.
        width_divisor = int(1 / width_downsample_ratio)
        # resize
        ratio = w / float(h)
        resize_w = math.ceil(imgH * ratio)
        if resize_w % width_divisor != 0:
            resize_w = round(resize_w / width_divisor) * width_divisor
        if imgW_min is not None:
            resize_w = max(imgW_min, resize_w)
        if imgW_max is not None:
            valid_ratio = min(1.0, 1.0 * resize_w / imgW_max)
            resize_w = min(imgW_max, resize_w)
        resized_image = cv2.resize(img, (resize_w, imgH))
        resized_image = resized_image.astype('float32')
        # norm 
        if image_shape[0] == 1:
            resized_image = resized_image / 255
            resized_image = resized_image[np.newaxis, :]
        else:
            resized_image = resized_image.transpose((2, 0, 1)) / 255
        resized_image -= 0.5
        resized_image /= 0.5
        resize_shape = resized_image.shape
        padding_im = -1.0 * np.ones((imgC, imgH, imgW_max), dtype=np.float32)
        padding_im[:, :, 0:resize_w] = resized_image
        pad_shape = padding_im.shape

        return padding_im, resize_shape, pad_shape, valid_ratio

L
LDOUBLEV 已提交
232 233
    def __call__(self, img_list):
        img_num = len(img_list)
234
        # Calculate the aspect ratio of all text bars
235 236 237
        width_list = []
        for img in img_list:
            width_list.append(img.shape[1] / float(img.shape[0]))
张欣-男's avatar
张欣-男 已提交
238
        # Sorting can speed up the recognition process
239 240
        indices = np.argsort(np.array(width_list))
        rec_res = [['', 0.0]] * img_num
241
        batch_num = self.rec_batch_num
L
LDOUBLEV 已提交
242
        st = time.time()
T
tink2123 已提交
243 244
        if self.benchmark:
            self.autolog.times.start()
L
LDOUBLEV 已提交
245 246 247
        for beg_img_no in range(0, img_num, batch_num):
            end_img_no = min(img_num, beg_img_no + batch_num)
            norm_img_batch = []
248
            max_wh_ratio = 0
L
LDOUBLEV 已提交
249
            for ino in range(beg_img_no, end_img_no):
250
                h, w = img_list[indices[ino]].shape[0:2]
251 252 253
                wh_ratio = w * 1.0 / h
                max_wh_ratio = max(max_wh_ratio, wh_ratio)
            for ino in range(beg_img_no, end_img_no):
T
Topdu 已提交
254
                if self.rec_algorithm != "SRN" and self.rec_algorithm != "SAR":
T
tink2123 已提交
255 256 257 258
                    norm_img = self.resize_norm_img(img_list[indices[ino]],
                                                    max_wh_ratio)
                    norm_img = norm_img[np.newaxis, :]
                    norm_img_batch.append(norm_img)
T
Topdu 已提交
259 260 261 262 263 264 265 266
                elif self.rec_algorithm == "SAR":
                    norm_img, _, _, valid_ratio = self.resize_norm_img_sar(
                        img_list[indices[ino]], self.rec_image_shape)
                    norm_img = norm_img[np.newaxis, :]
                    valid_ratio = np.expand_dims(valid_ratio, axis=0)
                    valid_ratios = []
                    valid_ratios.append(valid_ratio)
                    norm_img_batch.append(norm_img)
T
tink2123 已提交
267
                else:
L
LDOUBLEV 已提交
268 269
                    norm_img = self.process_image_srn(
                        img_list[indices[ino]], self.rec_image_shape, 8, 25)
T
tink2123 已提交
270 271 272 273 274 275 276 277 278
                    encoder_word_pos_list = []
                    gsrm_word_pos_list = []
                    gsrm_slf_attn_bias1_list = []
                    gsrm_slf_attn_bias2_list = []
                    encoder_word_pos_list.append(norm_img[1])
                    gsrm_word_pos_list.append(norm_img[2])
                    gsrm_slf_attn_bias1_list.append(norm_img[3])
                    gsrm_slf_attn_bias2_list.append(norm_img[4])
                    norm_img_batch.append(norm_img[0])
L
LDOUBLEV 已提交
279 280
            norm_img_batch = np.concatenate(norm_img_batch)
            norm_img_batch = norm_img_batch.copy()
T
tink2123 已提交
281 282
            if self.benchmark:
                self.autolog.times.stamp()
T
tink2123 已提交
283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308

            if self.rec_algorithm == "SRN":
                encoder_word_pos_list = np.concatenate(encoder_word_pos_list)
                gsrm_word_pos_list = np.concatenate(gsrm_word_pos_list)
                gsrm_slf_attn_bias1_list = np.concatenate(
                    gsrm_slf_attn_bias1_list)
                gsrm_slf_attn_bias2_list = np.concatenate(
                    gsrm_slf_attn_bias2_list)

                inputs = [
                    norm_img_batch,
                    encoder_word_pos_list,
                    gsrm_word_pos_list,
                    gsrm_slf_attn_bias1_list,
                    gsrm_slf_attn_bias2_list,
                ]
                input_names = self.predictor.get_input_names()
                for i in range(len(input_names)):
                    input_tensor = self.predictor.get_input_handle(input_names[
                        i])
                    input_tensor.copy_from_cpu(inputs[i])
                self.predictor.run()
                outputs = []
                for output_tensor in self.output_tensors:
                    output = output_tensor.copy_to_cpu()
                    outputs.append(output)
T
tink2123 已提交
309 310
                if self.benchmark:
                    self.autolog.times.stamp()
T
tink2123 已提交
311
                preds = {"predict": outputs[2]}
T
Topdu 已提交
312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330
            elif self.rec_algorithm == "SAR":
                valid_ratios = np.concatenate(valid_ratios)
                inputs = [
                    norm_img_batch,
                    valid_ratios,
                ]
                input_names = self.predictor.get_input_names()
                for i in range(len(input_names)):
                    input_tensor = self.predictor.get_input_handle(input_names[
                        i])
                    input_tensor.copy_from_cpu(inputs[i])
                self.predictor.run()
                outputs = []
                for output_tensor in self.output_tensors:
                    output = output_tensor.copy_to_cpu()
                    outputs.append(output)
                if self.benchmark:
                    self.autolog.times.stamp()
                preds = outputs[0]
T
tink2123 已提交
331 332 333 334 335 336 337
            else:
                self.input_tensor.copy_from_cpu(norm_img_batch)
                self.predictor.run()
                outputs = []
                for output_tensor in self.output_tensors:
                    output = output_tensor.copy_to_cpu()
                    outputs.append(output)
T
tink2123 已提交
338 339
                if self.benchmark:
                    self.autolog.times.stamp()
T
Topdu 已提交
340 341 342 343
                if len(outputs) != 1:
                    preds = outputs
                else:
                    preds = outputs[0]
W
WenmuZhou 已提交
344 345 346
            rec_result = self.postprocess_op(preds)
            for rno in range(len(rec_result)):
                rec_res[indices[beg_img_no + rno]] = rec_result[rno]
T
tink2123 已提交
347 348
            if self.benchmark:
                self.autolog.times.end(stamp=True)
L
LDOUBLEV 已提交
349
        return rec_res, time.time() - st
L
LDOUBLEV 已提交
350 351


352
def main(args):
D
dyning 已提交
353
    image_file_list = get_image_file_list(args.image_dir)
L
LDOUBLEV 已提交
354 355 356
    text_recognizer = TextRecognizer(args)
    valid_image_file_list = []
    img_list = []
L
LDOUBLEV 已提交
357

358
    # warmup 2 times
L
LDOUBLEV 已提交
359 360
    if args.warmup:
        img = np.random.uniform(0, 255, [32, 320, 3]).astype(np.uint8)
361
        for i in range(2):
L
LDOUBLEV 已提交
362
            res = text_recognizer([img] * int(args.rec_batch_num))
L
LDOUBLEV 已提交
363

L
LDOUBLEV 已提交
364
    for image_file in image_file_list:
L
LDOUBLEV 已提交
365 366 367
        img, flag = check_and_read_gif(image_file)
        if not flag:
            img = cv2.imread(image_file)
L
LDOUBLEV 已提交
368 369 370 371 372
        if img is None:
            logger.info("error in loading image:{}".format(image_file))
            continue
        valid_image_file_list.append(image_file)
        img_list.append(img)
L
LDOUBLEV 已提交
373 374 375 376 377 378 379 380 381 382
    try:
        rec_res, _ = text_recognizer(img_list)

    except Exception as E:
        logger.info(traceback.format_exc())
        logger.info(E)
        exit()
    for ino in range(len(img_list)):
        logger.info("Predicts of {}:{}".format(valid_image_file_list[ino],
                                               rec_res[ino]))
T
tink2123 已提交
383 384
    if args.benchmark:
        text_recognizer.autolog.report()
385 386 387 388


if __name__ == "__main__":
    main(utility.parse_args())