dataset_traversal.py 12.7 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 Reserve.
#
#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 os
tink2123's avatar
tink2123 committed
16
import sys
LDOUBLEV's avatar
LDOUBLEV committed
17
18
19
20
21
22
23
24
25
import math
import random
import numpy as np
import cv2

import string
import lmdb

from ppocr.utils.utility import initial_logger
tink2123's avatar
tink2123 committed
26
from ppocr.utils.utility import get_image_file_list
LDOUBLEV's avatar
LDOUBLEV committed
27
28
logger = initial_logger()

tink2123's avatar
tink2123 committed
29
from .img_tools import process_image, process_image_srn, get_img_data
LDOUBLEV's avatar
LDOUBLEV committed
30
31
32
33
34
35
36
37
38
39
40
41
42


class LMDBReader(object):
    def __init__(self, params):
        if params['mode'] != 'train':
            self.num_workers = 1
        else:
            self.num_workers = params['num_workers']
        self.lmdb_sets_dir = params['lmdb_sets_dir']
        self.char_ops = params['char_ops']
        self.image_shape = params['image_shape']
        self.loss_type = params['loss_type']
        self.max_text_length = params['max_text_length']
tink2123's avatar
tink2123 committed
43
        self.num_heads = params['num_heads']
LDOUBLEV's avatar
LDOUBLEV committed
44
        self.mode = params['mode']
tink2123's avatar
tink2123 committed
45
        self.drop_last = False
tink2123's avatar
tink2123 committed
46
        self.use_tps = False
tink2123's avatar
tink2123 committed
47
        if "tps" in params:
tink2123's avatar
tink2123 committed
48
            self.ues_tps = True
tink2123's avatar
tink2123 committed
49
        self.use_distort = False
tink2123's avatar
tink2123 committed
50
        if "distort" in params:
tink2123's avatar
tink2123 committed
51
52
53
54
55
            self.use_distort = params['distort'] and params['use_gpu']
            if not params['use_gpu']:
                logger.info(
                    "Distort operation can only support in GPU. Distort will be set to False."
                )
LDOUBLEV's avatar
LDOUBLEV committed
56
57
        if params['mode'] == 'train':
            self.batch_size = params['train_batch_size_per_card']
tink2123's avatar
tink2123 committed
58
            self.drop_last = True
tink2123's avatar
tink2123 committed
59
        else:
LDOUBLEV's avatar
LDOUBLEV committed
60
            self.batch_size = params['test_batch_size_per_card']
tink2123's avatar
tink2123 committed
61
            self.drop_last = False
62
            self.use_distort = False
tink2123's avatar
tink2123 committed
63
64
        self.infer_img = params['infer_img']

LDOUBLEV's avatar
LDOUBLEV committed
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
    def load_hierarchical_lmdb_dataset(self):
        lmdb_sets = {}
        dataset_idx = 0
        for dirpath, dirnames, filenames in os.walk(self.lmdb_sets_dir + '/'):
            if not dirnames:
                env = lmdb.open(
                    dirpath,
                    max_readers=32,
                    readonly=True,
                    lock=False,
                    readahead=False,
                    meminit=False)
                txn = env.begin(write=False)
                num_samples = int(txn.get('num-samples'.encode()))
                lmdb_sets[dataset_idx] = {"dirpath":dirpath, "env":env, \
                    "txn":txn, "num_samples":num_samples}
                dataset_idx += 1
        return lmdb_sets

    def print_lmdb_sets_info(self, lmdb_sets):
        lmdb_info_strs = []
        for dataset_idx in range(len(lmdb_sets)):
            tmp_str = " %s:%d," % (lmdb_sets[dataset_idx]['dirpath'],
                                   lmdb_sets[dataset_idx]['num_samples'])
            lmdb_info_strs.append(tmp_str)
        lmdb_info_strs = ''.join(lmdb_info_strs)
        logger.info("DataSummary:" + lmdb_info_strs)
        return

    def close_lmdb_dataset(self, lmdb_sets):
        for dataset_idx in lmdb_sets:
            lmdb_sets[dataset_idx]['env'].close()
        return

    def get_lmdb_sample_info(self, txn, index):
        label_key = 'label-%09d'.encode() % index
        label = txn.get(label_key)
        if label is None:
            return None
        label = label.decode('utf-8')
        img_key = 'image-%09d'.encode() % index
        imgbuf = txn.get(img_key)
        img = get_img_data(imgbuf)
        if img is None:
            return None
        return img, label

    def __call__(self, process_id):
        if self.mode != 'train':
            process_id = 0

        def sample_iter_reader():
tink2123's avatar
tink2123 committed
117
            if self.mode != 'train' and self.infer_img is not None:
tink2123's avatar
tink2123 committed
118
119
120
                image_file_list = get_image_file_list(self.infer_img)
                for single_img in image_file_list:
                    img = cv2.imread(single_img)
tink2123's avatar
tink2123 committed
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
                    if img.shape[-1]==1 or len(list(img.shape))==2:
                        img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
                    if self.loss_type == 'srn':
                        norm_img = process_image_srn(
                            img=img,
                            image_shape=self.image_shape,
                            num_heads=self.num_heads,
                            max_text_length=self.max_text_length
                        )
                    else:
                        norm_img = process_image(
                            img=img,
                            image_shape=self.image_shape,
                            char_ops=self.char_ops,
                            tps=self.use_tps,
                            infer_mode=True)
                    yield norm_img
            elif self.mode == 'test':
                image_file_list = get_image_file_list(self.infer_img)
                for single_img in image_file_list:
                    img = cv2.imread(single_img)
                    if img.shape[-1]==1 or len(list(img.shape))==2:
tink2123's avatar
tink2123 committed
143
                        img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
tink2123's avatar
tink2123 committed
144
145
146
                    norm_img = process_image(
                        img=img,
                        image_shape=self.image_shape,
tink2123's avatar
tink2123 committed
147
                        char_ops=self.char_ops,
tink2123's avatar
tink2123 committed
148
                        tps=self.use_tps,
tink2123's avatar
tink2123 committed
149
150
                        infer_mode=True
                    )
tink2123's avatar
tink2123 committed
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
                    yield norm_img
            else:
                lmdb_sets = self.load_hierarchical_lmdb_dataset()
                if process_id == 0:
                    self.print_lmdb_sets_info(lmdb_sets)
                cur_index_sets = [1 + process_id] * len(lmdb_sets)
                while True:
                    finish_read_num = 0
                    for dataset_idx in range(len(lmdb_sets)):
                        cur_index = cur_index_sets[dataset_idx]
                        if cur_index > lmdb_sets[dataset_idx]['num_samples']:
                            finish_read_num += 1
                        else:
                            sample_info = self.get_lmdb_sample_info(
                                lmdb_sets[dataset_idx]['txn'], cur_index)
                            cur_index_sets[dataset_idx] += self.num_workers
                            if sample_info is None:
                                continue
                            img, label = sample_info
tink2123's avatar
tink2123 committed
170
171
172
173
174
175
176
177
178
179
                            outs = []
                            if self.loss_type == "srn":
                                outs = process_image_srn(img, self.image_shape, self.num_heads,
                                                         self.max_text_length, label,
                                                         self.char_ops, self.loss_type)

                            else:
                                outs = process_image(img, self.image_shape, label,
                                                    self.char_ops, self.loss_type,
                                                    self.max_text_length)
tink2123's avatar
tink2123 committed
180
181
182
183
184
185
186
                            if outs is None:
                                continue
                            yield outs

                    if finish_read_num == len(lmdb_sets):
                        break
                self.close_lmdb_dataset(lmdb_sets)
LDOUBLEV's avatar
LDOUBLEV committed
187
188
189
190
191
192
193
        def batch_iter_reader():
            batch_outs = []
            for outs in sample_iter_reader():
                batch_outs.append(outs)
                if len(batch_outs) == self.batch_size:
                    yield batch_outs
                    batch_outs = []
tink2123's avatar
tink2123 committed
194
195
            if len(batch_outs) != 0:
                yield batch_outs
LDOUBLEV's avatar
LDOUBLEV committed
196

tink2123's avatar
tink2123 committed
197
        if self.infer_img is None:
tink2123's avatar
tink2123 committed
198
199
            return batch_iter_reader
        return sample_iter_reader
LDOUBLEV's avatar
LDOUBLEV committed
200
201
202
203
204
205
206
207


class SimpleReader(object):
    def __init__(self, params):
        if params['mode'] != 'train':
            self.num_workers = 1
        else:
            self.num_workers = params['num_workers']
tink2123's avatar
tink2123 committed
208
209
210
        if params['mode'] != 'test':
            self.img_set_dir = params['img_set_dir']
            self.label_file_path = params['label_file_path']
tink2123's avatar
tink2123 committed
211
        self.use_gpu = params['use_gpu']
LDOUBLEV's avatar
LDOUBLEV committed
212
213
214
215
216
        self.char_ops = params['char_ops']
        self.image_shape = params['image_shape']
        self.loss_type = params['loss_type']
        self.max_text_length = params['max_text_length']
        self.mode = params['mode']
tink2123's avatar
tink2123 committed
217
        self.infer_img = params['infer_img']
tink2123's avatar
tink2123 committed
218
        self.use_tps = False
tink2123's avatar
tink2123 committed
219
        if "tps" in params:
tink2123's avatar
tink2123 committed
220
            self.use_tps = True
tink2123's avatar
tink2123 committed
221
        self.use_distort = False
tink2123's avatar
tink2123 committed
222
        if "distort" in params:
tink2123's avatar
tink2123 committed
223
224
225
226
227
            self.use_distort = params['distort'] and params['use_gpu']
            if not params['use_gpu']:
                logger.info(
                    "Distort operation can only support in GPU.Distort will be set to False."
                )
LDOUBLEV's avatar
LDOUBLEV committed
228
229
        if params['mode'] == 'train':
            self.batch_size = params['train_batch_size_per_card']
tink2123's avatar
tink2123 committed
230
            self.drop_last = True
LDOUBLEV's avatar
LDOUBLEV committed
231
        else:
tink2123's avatar
tink2123 committed
232
            self.batch_size = params['test_batch_size_per_card']
tink2123's avatar
tink2123 committed
233
            self.drop_last = False
234
            self.use_distort = False
LDOUBLEV's avatar
LDOUBLEV committed
235
236
237
238
239

    def __call__(self, process_id):
        if self.mode != 'train':
            process_id = 0

tink2123's avatar
tink2123 committed
240
241
242
243
244
245
246
247
248
        def get_device_num():
            if self.use_gpu:
                gpus = os.environ.get("CUDA_VISIBLE_DEVICES", 1)
                gpu_num = len(gpus.split(','))
                return gpu_num
            else:
                cpu_num = os.environ.get("CPU_NUM", 1)
                return int(cpu_num)

LDOUBLEV's avatar
LDOUBLEV committed
249
        def sample_iter_reader():
tink2123's avatar
tink2123 committed
250
            if self.mode != 'train' and self.infer_img is not None:
tink2123's avatar
tink2123 committed
251
                image_file_list = get_image_file_list(self.infer_img)
tink2123's avatar
tink2123 committed
252
253
                for single_img in image_file_list:
                    img = cv2.imread(single_img)
tink2123's avatar
tink2123 committed
254
                    if img.shape[-1] == 1 or len(list(img.shape)) == 2:
tink2123's avatar
tink2123 committed
255
                        img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
tink2123's avatar
tink2123 committed
256
257
258
                    norm_img = process_image(
                        img=img,
                        image_shape=self.image_shape,
tink2123's avatar
tink2123 committed
259
                        char_ops=self.char_ops,
tink2123's avatar
tink2123 committed
260
                        tps=self.use_tps,
tink2123's avatar
tink2123 committed
261
                        infer_mode=True)
tink2123's avatar
tink2123 committed
262
                    yield norm_img
tink2123's avatar
tink2123 committed
263
264
265
266
267
268
            else:
                with open(self.label_file_path, "rb") as fin:
                    label_infor_list = fin.readlines()
                img_num = len(label_infor_list)
                img_id_list = list(range(img_num))
                random.shuffle(img_id_list)
littletomatodonkey's avatar
littletomatodonkey committed
269
                if sys.platform == "win32" and self.num_workers != 1:
tink2123's avatar
tink2123 committed
270
271
272
                    print("multiprocess is not fully compatible with Windows."
                          "num_workers will be 1.")
                    self.num_workers = 1
tink2123's avatar
tink2123 committed
273
                if self.batch_size * get_device_num() > img_num:
tink2123's avatar
tink2123 committed
274
                    raise Exception(
tink2123's avatar
tink2123 committed
275
276
                        "The number of the whole data ({}) is smaller than the batch_size * devices_num ({})".
                        format(img_num, self.batch_size * get_device_num()))
tink2123's avatar
tink2123 committed
277
278
279
280
281
282
283
284
                for img_id in range(process_id, img_num, self.num_workers):
                    label_infor = label_infor_list[img_id_list[img_id]]
                    substr = label_infor.decode('utf-8').strip("\n").split("\t")
                    img_path = self.img_set_dir + "/" + substr[0]
                    img = cv2.imread(img_path)
                    if img is None:
                        logger.info("{} does not exist!".format(img_path))
                        continue
tink2123's avatar
tink2123 committed
285
                    if img.shape[-1] == 1 or len(list(img.shape)) == 2:
tink2123's avatar
tink2123 committed
286
287
                        img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)

tink2123's avatar
tink2123 committed
288
                    label = substr[1]
tink2123's avatar
tink2123 committed
289
290
291
292
293
294
295
296
                    outs = process_image(
                        img=img,
                        image_shape=self.image_shape,
                        label=label,
                        char_ops=self.char_ops,
                        loss_type=self.loss_type,
                        max_text_length=self.max_text_length,
                        distort=self.use_distort)
tink2123's avatar
tink2123 committed
297
298
299
                    if outs is None:
                        continue
                    yield outs
LDOUBLEV's avatar
LDOUBLEV committed
300
301
302
303
304
305
306
307

        def batch_iter_reader():
            batch_outs = []
            for outs in sample_iter_reader():
                batch_outs.append(outs)
                if len(batch_outs) == self.batch_size:
                    yield batch_outs
                    batch_outs = []
tink2123's avatar
tink2123 committed
308
309
310
            if not self.drop_last:
                if len(batch_outs) != 0:
                    yield batch_outs
LDOUBLEV's avatar
LDOUBLEV committed
311

tink2123's avatar
tink2123 committed
312
        if self.infer_img is None:
tink2123's avatar
tink2123 committed
313
            return batch_iter_reader
tink2123's avatar
tink2123 committed
314
        return sample_iter_reader