-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathquick_seg_label.py
165 lines (139 loc) · 6.71 KB
/
quick_seg_label.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
import argparse
import json
import os
from typing import Tuple
import cv2
from tqdm import tqdm
from object_detector.detector import ObjectDetector
from segment.segment_ai import SegmentAnythingAI
obj_detector = ObjectDetector()
segment_ai = SegmentAnythingAI(r'segment/segment_anything/sam_vit_h_4b8939.pth')
def pipeline(img_path: str, dsize: Tuple[int, int], save_img_dir: str, text_prompt: str, overlap: float):
if not os.path.exists(save_img_dir):
os.mkdir(save_img_dir)
filename = os.path.basename(img_path).split('.')[0]
filetype = os.path.basename(img_path).split('.')[1]
# 切成 4 块并保存到新文件夹
patches, _, _ = _patchify(cv2.imread(img_path), dsize, overlap=overlap, return_list=True)
img_list = []
for i, patch in enumerate(patches):
img_save_name = os.path.join(save_img_dir, f'{filename}_{i}.{filetype}')
cv2.imwrite(img_save_name, patch)
img_list.append(img_save_name)
for patch_path in img_list:
object_id = 0
img_anno = {'file_name': patch_path}
obj_anns = []
xyxy_list, _ = obj_detector.detect(img_path=patch_path,
text_prompt=text_prompt)
segment_ai.set_img(patch_path)
for xyxy in xyxy_list:
x1, y1, x2, y2 = xyxy
h = y2 - y1
if h > 2000:
continue # 跳过明显预测错误的框
# 因为 Sam 模型会切割出一些小点点出来，所有要找最大面积的 contour 所以只能一个一个找
points = segment_ai.detect_by_boxes(boxes=[xyxy, ])[0]
obj_anns.append({"obj_id": object_id,
"annotation": {"box": xyxy,
"category": text_prompt,
"mask": points}})
object_id += 1
img_anno['obj_anns'] = obj_anns
if not os.path.exists(os.path.join(save_img_dir, 'json')):
os.mkdir(os.path.join(save_img_dir, 'json'))
short_filename = os.path.basename(patch_path).split('.')[0]
with open(os.path.join(save_img_dir, 'json', f'{short_filename}.json'), 'w', encoding='utf8') as f:
json.dump(img_anno, f)
def pipeline_sam_only(img_path: str, dsize: Tuple[int, int], save_img_dir: str, text_prompt: str, overlap: float):
if not os.path.exists(save_img_dir):
os.mkdir(save_img_dir)
filename = os.path.basename(img_path).split('.')[0]
filetype = os.path.basename(img_path).split('.')[1]
# 切成 4 块并保存到新文件夹
# patches, _, _ = _patchify(cv2.imdecode(img_path), dsize, overlap=overlap, return_list=True)
patches, _, _ = _patchify(cv2.imread(img_path), dsize, overlap=overlap, return_list=True)
img_list = []
for i, patch in enumerate(patches):
img_save_name = os.path.join(save_img_dir, f'{filename}_{i}.{filetype}')
cv2.imwrite(img_save_name, patch)
img_list.append(img_save_name)
for patch_path in tqdm(img_list):
object_id = 0
img_anno = {'file_name': patch_path}
segment_ai.set_img(patch_path)
img_anno['obj_anns'] = segment_ai.detect_auto(patch_path, category=text_prompt, obj_id=object_id)
object_id += len(img_anno['obj_anns'])
if not os.path.exists(os.path.join(save_img_dir, 'json')):
os.mkdir(os.path.join(save_img_dir, 'json'))
short_filename = os.path.basename(patch_path).split('.')[0]
with open(os.path.join(save_img_dir, 'json', f'{short_filename}.json'), 'w', encoding='utf8') as f:
json.dump(img_anno, f)
def _patchify(img, dsize, overlap=0.2, return_list=False):
"""
将图片切割为 MxN 块 patches
:param img: numpy (h, w, channel)
:param dsize: (int_h, int_w)
:param overlap: float 各个 patch 之间重叠的像素比例
:param return_list: bool 如果为真，以数组的形式返回
:return: dict，key 是位置元组，value 是切割后的 numpy 数组
patches: dict[(i, j)]=crop_image
overlap_rate: dict[(i, j)]=((croped_h-step_h)/croped_h, (croped_w-step_w)/croped_w) 重叠的像素占 patch 的比例
start_pos: dict[(i, j)]=(h1, w1) patch左上角在原图中的坐标
"""
h, w = img.shape[:2]
num_h, num_w = dsize
step_h = int(h / num_h)
step_w = int(w / num_w)
overlap_h = int(h * overlap)
overlap_w = int(w * overlap)
patches = [] if return_list else {}
overlap_rate = [] if return_list else {}
start_pos = [] if return_list else {}
for i in range(num_h):
for j in range(num_w):
h1 = max(0, i * step_h - overlap_h)
w1 = max(0, j * step_w - overlap_w)
croped = img[h1:(i + 1) * step_h, w1:(j + 1) * step_w]
if return_list:
patches.append(croped)
else:
patches[(i, j)] = croped
croped_h, croped_w = croped.shape[:2]
overlap_r = ((croped_h - step_h) / croped_h, (croped_w - step_w) / croped_w)
if return_list:
overlap_rate.append(overlap_r)
else:
overlap_rate[(i, j)] = overlap_r
if return_list:
start_pos.append((h1, w1))
else:
start_pos[(i, j)] = (h1, w1)
return patches, overlap_rate, start_pos
if __name__ == '__main__':
# parser = argparse.ArgumentParser()
# parser.add_argument('--sr_dir', type=str, help='The root path of images to be detected')
# parser.add_argument('--ds_dir', type=str, help='The root path of patches and json files to be saved')
# parser.add_argument('--prompt', type=str, help='The prompt')
# args = parser.parse_args()
src_root_dir = 'G:\weeds_2024\Shanghai_mix'
dst_root_dir = 'G:\weeds_2024\Shanghai_mix_processing'
subdirs = os.listdir(src_root_dir)
for subdir in subdirs:
print('='*50, subdir, '='*50)
if not os.path.exists(os.path.join(dst_root_dir, subdir)):
os.mkdir(os.path.join(dst_root_dir, subdir))
args = {
'sr_dir': os.path.join(src_root_dir, subdir),
'ds_dir': os.path.join(dst_root_dir, subdir),
'prompt': 'wheat'
}
files = os.listdir(args['sr_dir'])
for img in tqdm(files):
# pipeline_sam_only(os.path.join(args.sr_dir, img), (3, 3), args.ds_dir, args.prompt, overlap=0.2)
try:
# pipeline_sam_only(os.path.join(args.sr_dir, img), (3, 3), args.ds_dir, args.prompt, overlap=0.2)
pipeline(os.path.join(args['sr_dir'], img), (3, 3), args['ds_dir'], args['prompt'], overlap=0.0)
except Exception as e:
print(os.path.join(args['sr_dir'], img), 'can not be detected')
print(e)