forked from mindspore-lab/mindocr
-
Notifications
You must be signed in to change notification settings - Fork 0
/
synthtext.py
175 lines (150 loc) · 6.8 KB
/
synthtext.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
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
117
118
119
120
121
122
123
124
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
import itertools
import os
from collections import defaultdict
from concurrent.futures import ProcessPoolExecutor
from typing import Any, Dict, Tuple
import cv2
import numpy as np
from PIL import Image
from scipy.io import loadmat, savemat
from shapely.geometry import Polygon, box
from tqdm import tqdm
from mindocr.data.utils.polygon_utils import sort_clockwise
from tools.dataset_converters.utils.lmdb_writer import create_lmdb_dataset
from tools.infer.text.utils import crop_text_region
class SYNTHTEXT_Converter:
"""
Validate polygons and sort vertices in SynthText dataset. The filtered dataset will be stored
in the same format as the original one for compatibility purposes.
Args:
min_area: area below which polygons will be filtered out
"""
def __init__(self, *args, **kwargs):
self._image_dir = None
def _sort_and_validate(self, sample: Tuple[np.ndarray, ...]) -> Tuple[np.ndarray, ...]:
"""
Sort vertices in clockwise order (to eliminate self-intersection) and filter invalid polygons out.
Args:
sample: tuple containing polygons and texts instances.
Returns:
filtered polygons and texts.
"""
path, polys, texts = sample
polys = polys.transpose().reshape(-1, 4, 2) # some labels have (4, 2) shape (no batch dimension)
texts = [t for text in texts.tolist() for t in text.split()] # TODO: check the correctness of texts order
size = np.array(Image.open(os.path.join(self._image_dir, path.item())).size) - 1 # (w, h)
border = box(0, 0, *size)
# SynthText has a lot of mistakes in the dataset that may affect the data processing pipeline
# Sort vertices clockwise and filter invalid polygons out
new_polys, new_texts = [], []
for np_poly, text in zip(polys, texts):
# fix self-intersection by sorting vertices
np_poly = sort_clockwise(np_poly)
# check if the polygon is valid and lies within the visible borders
poly = Polygon(np_poly)
if poly.is_valid and poly.intersects(border):
np_poly = np.clip(np_poly, 0, size) # clip bbox to be within the visible region
poly = Polygon(np_poly) # check the polygon validity once again after clipping
if poly.is_valid and not poly.equals(border):
new_polys.append(np_poly)
new_texts.append(text)
return np.array(new_polys).transpose(), np.array(new_texts) # preserve polygons' axes order
def convert(self, task="det", image_dir=None, label_path=None, output_path=None):
if task == "det":
self.convert_det(image_dir, label_path, output_path, save_output=True)
elif task == "rec_lmdb":
self.convert_rec_lmdb(image_dir, label_path, output_path)
else:
raise ValueError(f"Unsupported task `{task}`.")
def convert_det(self, image_dir=None, label_path=None, output_path=None):
self._image_dir = image_dir
print("Loading SynthText dataset. It might take a while...")
mat = loadmat(label_path)
# use multiprocessing to process the dataset faster
with ProcessPoolExecutor(max_workers=8) as pool:
data_list = list(
tqdm(
pool.map(self._sort_and_validate, zip(mat["imnames"][0], mat["wordBB"][0], mat["txt"][0])),
total=len(mat["imnames"][0]),
desc="Processing data",
miniters=10000,
)
)
wordBB, txt = zip(*data_list)
for i in range(len(mat["wordBB"][0])): # how to stack wordBB?
mat["wordBB"][0][i] = wordBB[i]
mat["txt"] = np.array(txt).reshape(1, -1)
print("Saving...")
savemat(
output_path,
{
"charBB": mat["charBB"], # save as it is
"wordBB": mat["wordBB"],
"imnames": mat["imnames"],
"txt": mat["txt"],
},
do_compression=True,
)
def _crop_with_single_text(self, sample: Tuple[str, Dict[str, Any]]) -> Tuple[bytes, str]:
images = []
labels = []
img_path, img_info = sample
try:
img = cv2.imread(img_path, cv2.IMREAD_COLOR)
except Exception as e:
print(f"WARNING: {str(e)}")
return images, labels
for item_info in img_info:
try:
sub_img = crop_text_region(img, item_info["poly"], box_type="poly", rotate_if_vertical=False)
# encode image as JPEG format
sub_img = cv2.imencode(".jpg", sub_img)[1].tobytes()
except Exception as e:
print(f"WARNING: {str(e)}")
continue
sub_text = item_info["text"]
images.append(sub_img)
labels.append(sub_text)
return images, labels
def convert_rec_lmdb(self, image_dir=None, label_path=None, output_path=None):
self._image_dir = image_dir
print("Loading SynthText dataset. It might take a while...")
mat = loadmat(label_path)
# use multiprocessing to process the dataset faster
with ProcessPoolExecutor(max_workers=8) as pool:
data_list = list(
tqdm(
pool.map(self._sort_and_validate, zip(mat["imnames"][0], mat["wordBB"][0], mat["txt"][0])),
total=len(mat["imnames"][0]),
desc="Processing data",
)
)
wordBB, txt = zip(*data_list)
for i in range(len(mat["wordBB"][0])):
mat["wordBB"][0][i] = wordBB[i]
mat["txt"] = np.array(txt).reshape(1, -1)
data_list = defaultdict(list)
for image, polys, texts in zip(mat["imnames"][0], mat["wordBB"][0], mat["txt"][0]):
texts = [t for text in texts.tolist() for t in text.split()]
polys = polys.transpose().reshape(-1, 4, 2)
img_path = os.path.join(image_dir, image.item())
for poly, text in zip(polys, texts):
data_list[img_path].append(
{
"poly": poly,
"text": text,
}
)
with ProcessPoolExecutor(max_workers=8) as pool:
data_list = list(
tqdm(
pool.map(self._crop_with_single_text, zip(data_list.keys(), data_list.values())),
total=len(data_list),
desc="Cropping data",
)
)
images, labels = zip(*data_list)
images = iter(itertools.chain(*images))
labels = iter(itertools.chain(*labels))
print("Creating the LMDB dataset.")
create_lmdb_dataset(images, labels, output_path=output_path)