Doraemon-AI
commited on
up
Browse files- labelme2coco.py +187 -0
- labels.txt +12 -0
labelme2coco.py
ADDED
@@ -0,0 +1,187 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/usr/bin/env python
|
2 |
+
|
3 |
+
import argparse
|
4 |
+
import collections
|
5 |
+
import datetime
|
6 |
+
import glob
|
7 |
+
import json
|
8 |
+
import os
|
9 |
+
import os.path as osp
|
10 |
+
import sys
|
11 |
+
import uuid
|
12 |
+
|
13 |
+
import imgviz
|
14 |
+
import numpy as np
|
15 |
+
|
16 |
+
import labelme
|
17 |
+
|
18 |
+
try:
|
19 |
+
import pycocotools.mask
|
20 |
+
except ImportError:
|
21 |
+
print("Please install pycocotools:\n\n pip install pycocotools\n")
|
22 |
+
sys.exit(1)
|
23 |
+
|
24 |
+
|
25 |
+
def main():
|
26 |
+
parser = argparse.ArgumentParser(
|
27 |
+
formatter_class=argparse.ArgumentDefaultsHelpFormatter
|
28 |
+
)
|
29 |
+
parser.add_argument("input_dir", help="input annotated directory")
|
30 |
+
parser.add_argument("output_dir", help="output dataset directory")
|
31 |
+
parser.add_argument("--labels", help="labels file", required=True)
|
32 |
+
parser.add_argument(
|
33 |
+
"--noviz", help="no visualization", action="store_true"
|
34 |
+
)
|
35 |
+
args = parser.parse_args()
|
36 |
+
|
37 |
+
if osp.exists(args.output_dir):
|
38 |
+
print("Output directory already exists:", args.output_dir)
|
39 |
+
sys.exit(1)
|
40 |
+
os.makedirs(args.output_dir)
|
41 |
+
os.makedirs(osp.join(args.output_dir, "JPEGImages"))
|
42 |
+
if not args.noviz:
|
43 |
+
os.makedirs(osp.join(args.output_dir, "Visualization"))
|
44 |
+
print("Creating dataset:", args.output_dir)
|
45 |
+
|
46 |
+
now = datetime.datetime.now()
|
47 |
+
|
48 |
+
data = dict(
|
49 |
+
info=dict(
|
50 |
+
description=None,
|
51 |
+
url=None,
|
52 |
+
version=None,
|
53 |
+
year=now.year,
|
54 |
+
contributor=None,
|
55 |
+
date_created=now.strftime("%Y-%m-%d %H:%M:%S.%f"),
|
56 |
+
),
|
57 |
+
licenses=[dict(url=None, id=0, name=None,)],
|
58 |
+
images=[
|
59 |
+
# license, url, file_name, height, width, date_captured, id
|
60 |
+
],
|
61 |
+
type="instances",
|
62 |
+
annotations=[
|
63 |
+
# segmentation, area, iscrowd, image_id, bbox, category_id, id
|
64 |
+
],
|
65 |
+
categories=[
|
66 |
+
# supercategory, id, name
|
67 |
+
],
|
68 |
+
)
|
69 |
+
|
70 |
+
class_name_to_id = {}
|
71 |
+
for i, line in enumerate(open(args.labels).readlines()):
|
72 |
+
class_id = i - 1 # starts with -1
|
73 |
+
class_name = line.strip()
|
74 |
+
if class_id == -1:
|
75 |
+
assert class_name == "__ignore__"
|
76 |
+
continue
|
77 |
+
class_name_to_id[class_name] = class_id
|
78 |
+
data["categories"].append(
|
79 |
+
dict(supercategory=None, id=class_id, name=class_name,)
|
80 |
+
)
|
81 |
+
|
82 |
+
out_ann_file = osp.join(args.output_dir, "annotations.json")
|
83 |
+
label_files = glob.glob(osp.join(args.input_dir, "*.json"))
|
84 |
+
for image_id, filename in enumerate(label_files):
|
85 |
+
print("Generating dataset from:", filename)
|
86 |
+
|
87 |
+
label_file = labelme.LabelFile(filename=filename)
|
88 |
+
|
89 |
+
base = osp.splitext(osp.basename(filename))[0]
|
90 |
+
out_img_file = osp.join(args.output_dir, "JPEGImages", base + ".jpg")
|
91 |
+
|
92 |
+
img = labelme.utils.img_data_to_arr(label_file.imageData)
|
93 |
+
imgviz.io.imsave(out_img_file, img)
|
94 |
+
data["images"].append(
|
95 |
+
dict(
|
96 |
+
license=0,
|
97 |
+
url=None,
|
98 |
+
file_name=osp.relpath(out_img_file, osp.dirname(out_ann_file)),
|
99 |
+
height=img.shape[0],
|
100 |
+
width=img.shape[1],
|
101 |
+
date_captured=None,
|
102 |
+
id=image_id,
|
103 |
+
)
|
104 |
+
)
|
105 |
+
|
106 |
+
masks = {} # for area
|
107 |
+
segmentations = collections.defaultdict(list) # for segmentation
|
108 |
+
for shape in label_file.shapes:
|
109 |
+
points = shape["points"]
|
110 |
+
label = shape["label"]
|
111 |
+
group_id = shape.get("group_id")
|
112 |
+
shape_type = shape.get("shape_type", "polygon")
|
113 |
+
mask = labelme.utils.shape_to_mask(
|
114 |
+
img.shape[:2], points, shape_type
|
115 |
+
)
|
116 |
+
|
117 |
+
if group_id is None:
|
118 |
+
group_id = uuid.uuid1()
|
119 |
+
|
120 |
+
instance = (label, group_id)
|
121 |
+
|
122 |
+
if instance in masks:
|
123 |
+
masks[instance] = masks[instance] | mask
|
124 |
+
else:
|
125 |
+
masks[instance] = mask
|
126 |
+
|
127 |
+
if shape_type == "rectangle":
|
128 |
+
(x1, y1), (x2, y2) = points
|
129 |
+
x1, x2 = sorted([x1, x2])
|
130 |
+
y1, y2 = sorted([y1, y2])
|
131 |
+
points = [x1, y1, x2, y1, x2, y2, x1, y2]
|
132 |
+
else:
|
133 |
+
points = np.asarray(points).flatten().tolist()
|
134 |
+
|
135 |
+
segmentations[instance].append(points)
|
136 |
+
segmentations = dict(segmentations)
|
137 |
+
|
138 |
+
for instance, mask in masks.items():
|
139 |
+
cls_name, group_id = instance
|
140 |
+
if cls_name not in class_name_to_id:
|
141 |
+
continue
|
142 |
+
cls_id = class_name_to_id[cls_name]
|
143 |
+
|
144 |
+
mask = np.asfortranarray(mask.astype(np.uint8))
|
145 |
+
mask = pycocotools.mask.encode(mask)
|
146 |
+
area = float(pycocotools.mask.area(mask))
|
147 |
+
bbox = pycocotools.mask.toBbox(mask).flatten().tolist()
|
148 |
+
|
149 |
+
data["annotations"].append(
|
150 |
+
dict(
|
151 |
+
id=len(data["annotations"]),
|
152 |
+
image_id=image_id,
|
153 |
+
category_id=cls_id,
|
154 |
+
segmentation=segmentations[instance],
|
155 |
+
area=area,
|
156 |
+
bbox=bbox,
|
157 |
+
iscrowd=0,
|
158 |
+
)
|
159 |
+
)
|
160 |
+
|
161 |
+
if not args.noviz:
|
162 |
+
labels, captions, masks = zip(
|
163 |
+
*[
|
164 |
+
(class_name_to_id[cnm], cnm, msk)
|
165 |
+
for (cnm, gid), msk in masks.items()
|
166 |
+
if cnm in class_name_to_id
|
167 |
+
]
|
168 |
+
)
|
169 |
+
viz = imgviz.instances2rgb(
|
170 |
+
image=img,
|
171 |
+
labels=labels,
|
172 |
+
masks=masks,
|
173 |
+
captions=captions,
|
174 |
+
font_size=15,
|
175 |
+
line_width=2,
|
176 |
+
)
|
177 |
+
out_viz_file = osp.join(
|
178 |
+
args.output_dir, "Visualization", base + ".jpg"
|
179 |
+
)
|
180 |
+
imgviz.io.imsave(out_viz_file, viz)
|
181 |
+
|
182 |
+
with open(out_ann_file, "w") as f:
|
183 |
+
json.dump(data, f)
|
184 |
+
|
185 |
+
|
186 |
+
if __name__ == "__main__":
|
187 |
+
main()
|
labels.txt
ADDED
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
__ignore__
|
2 |
+
_background_
|
3 |
+
Text
|
4 |
+
Title
|
5 |
+
Figure
|
6 |
+
Figure caption
|
7 |
+
Table
|
8 |
+
Table caption
|
9 |
+
Header
|
10 |
+
Footer
|
11 |
+
Reference
|
12 |
+
Equation
|