/
tri_dataset.py
308 lines (287 loc) · 13.4 KB
/
tri_dataset.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
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
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
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
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
from typing import Callable, Optional
from torchvision.datasets import VisionDataset
from PIL import Image
import os
from typing import Any, Callable, cast, Dict, List, Optional, Tuple
import numpy as np
import albumentations as A
from albumentations.pytorch import ToTensorV2
import torch
import cv2
def transform():
return A.Compose([
A.Resize(720,1280),
A.Normalize(mean=[0.485, 0.456, 0.406],std=[0.229, 0.224, 0.225]),
ToTensorV2(),
])
def video_transform():
return A.Compose([
A.Normalize(mean=[0.485, 0.456, 0.406],std=[0.229, 0.224, 0.225]),
ToTensorV2(),
])
def inverse_normalize(img):
img=img*torch.tensor([0.229, 0.224, 0.225]).view(3,1,1)
img=img+torch.tensor([0.485, 0.456, 0.406]).view(3,1,1)
img=img*255
return img
class VideoDataset(VisionDataset):
def __init__(self, root: str, annotation_dir:str,transforms: Callable[..., Any] | None = None, transform: Callable[..., Any] | None = None, target_transform: Callable[..., Any] | None = None) -> None:
super().__init__(root, transforms, transform, target_transform)
self.video_paths=os.listdir(root)
self.video_paths=[os.path.join(root,video_path) for video_path in self.video_paths]
self.annotations_paths=os.listdir(annotation_dir)
self.annotations_paths=[os.path.join(annotation_dir,annotation) for annotation in self.annotations_paths]
self.annotation={}
self.videos=[]
for annotation_path in self.annotations_paths:
with open(annotation_path) as f:
video_name=annotation_path.split('/')[-1].split('.')[0]
self.annotation[video_name]=[]
lines=f.readlines()
for line in lines:
line=line.split()
file_name=line[0]
width,height=int(line[1]),int(line[2])
data=[float(x) for x in line[3:] if x!=""]
cnt=len(data)//7
if len(data)>0:
ids=-1*np.ones((cnt,1))
pts=-1*np.ones((cnt,2))
bboxes=-1*np.ones((cnt,4))
data=np.array(data)
data=np.reshape(data,(-1,7))
ids=data[:,6].reshape(-1,1)
pts=data[:,4:6]
bboxes[:]=data[:,0:4]
bboxes[:,2]=bboxes[:,2]-bboxes[:,0]
bboxes[:,3]=bboxes[:,3]-bboxes[:,1]
pts[:,0]=pts[:,0]/width
pts[:,1]=pts[:,1]/height
bboxes[:,0]=bboxes[:,0]/width
bboxes[:,1]=bboxes[:,1]/height
bboxes[:,2]=bboxes[:,2]/width
bboxes[:,3]=bboxes[:,3]/height
else:
ids=-1*np.ones((1,1))
pts=-1*np.ones((1,2))
bboxes=-1*np.ones((1,4))
self.annotation[video_name].append({"file_name":file_name,"height":height,"width":width,"ids":ids,"pts":pts,"bboxes":bboxes,"cnt":cnt})
for video_path in self.video_paths:
video_name=video_path.split('/')[-1]
self.videos.append({
"video_name":video_name,
"img_names":[],
"height":self.annotation[video_name][0]["height"],
"width":self.annotation[video_name][0]["width"],
"ids":[],
"pts":[],
"bboxes":[],
"cnt":[],
})
for i in range(0,len(self.annotation[video_name])):
self.videos[-1]["img_names"].append(self.annotation[video_name][i]["file_name"])
self.videos[-1]["ids"].append(self.annotation[video_name][i]["ids"])
self.videos[-1]["pts"].append(self.annotation[video_name][i]["pts"])
self.videos[-1]["bboxes"].append(self.annotation[video_name][i]["bboxes"])
self.videos[-1]["cnt"].append(self.annotation[video_name][i]["cnt"])
def __len__(self) -> int:
return len(self.videos)
def __getitem__(self, index: int) -> Tuple[Any, Any]:
video=self.videos[index]
video_name=video["video_name"]
img_names=video["img_names"]
height=video["height"]
width=video["width"]
ids=video["ids"]
pts=video["pts"]
bboxes=video["bboxes"]
cnt=video["cnt"]
imgs=[]
for img_name in img_names:
img_path=os.path.join(self.root,video_name,img_name)
img=self.transforms(image=np.array(Image.open(img_path).convert("RGB")))["image"]
imgs.append(img)
imgs=torch.stack(imgs,dim=0)
labels={
"h":height,
"w":width,
"pts":pts,
"bboxes":bboxes,
"cnt":cnt,
"video_name":video_name,
"img_names":img_names,
}
return imgs,labels
class PairDataset(VisionDataset):
def __init__(self, root: str, annotation_dir:str, max_len:int ,transforms: Callable[..., Any] | None = None, transform: Callable[..., Any] | None = None, target_transform: Callable[..., Any] | None = None, train=True,step=20,interval=1,force_last=False) -> None:
super().__init__(root, transforms, transform, target_transform)
self.video_paths=os.listdir(root)
self.video_paths=[os.path.join(root,video_path) for video_path in self.video_paths]
self.annotations_paths=os.listdir(annotation_dir)
self.annotations_paths=[os.path.join(annotation_dir,annotation) for annotation in self.annotations_paths]
self.annotation={}
self.pairs=[]
self.max_len=max_len
for annotation_path in self.annotations_paths:
with open(annotation_path) as f:
video_name=annotation_path.split('/')[-1].split('.')[0]
self.annotation[video_name]=[]
lines=f.readlines()
for line in lines:
line=line.split()
file_name=line[0]
width,height=int(line[1]),int(line[2])
data=[float(x) for x in line[3:] if x!=""]
cnt=len(data)//7
ids=-1*np.ones((max_len,1))
pts=-1*np.ones((max_len,2))
bboxes=-1*np.ones((max_len,4))
if len(data)>0:
ids=-1*np.ones((cnt,1))
pts=-1*np.ones((cnt,2))
bboxes=-1*np.ones((cnt,4))
data=np.array(data)
data=np.reshape(data,(-1,7))
ids=data[:,6].reshape(-1,1)
pts=data[:,4:6]
bboxes[:]=data[:,0:4]
bboxes[:,2]=bboxes[:,2]-bboxes[:,0]
bboxes[:,3]=bboxes[:,3]-bboxes[:,1]
pts[:,0]=pts[:,0]/width
pts[:,1]=pts[:,1]/height
bboxes[:,0]=bboxes[:,0]/width
bboxes[:,1]=bboxes[:,1]/height
bboxes[:,2]=bboxes[:,2]/width
bboxes[:,3]=bboxes[:,3]/height
self.annotation[video_name].append({"file_name":file_name,"height":height,"width":width,"ids":ids,"pts":pts,"bboxes":bboxes,"cnt":cnt})
for video_path in self.video_paths:
video_name=video_path.split('/')[-1]
last_step=0
for i in range(1,len(self.annotation[video_name])-step,interval):
self.pairs.append({
"0":self.annotation[video_name][i],
"1":self.annotation[video_name][i+step],
"video_name":video_name,
})
last_step=i+step
if force_last and last_step<len(self.annotation[video_name])-1:
self.pairs.append({
"0":self.annotation[video_name][last_step],
"1":self.annotation[video_name][-1],
"video_name":video_name,
})
self.train=train
def __len__(self) -> int:
return len(self.pairs)
def add_noise(self,pts):
noise=np.random.normal(scale=0.001,size=pts.shape)
pts=pts+noise
pts[pts>1]=1
pts[pts<0]=0
return pts
def __getitem__(self, index: int) -> Tuple[Any, Any]:
pair=self.pairs[index]
img0_path=os.path.join(self.root,pair["video_name"],pair["0"]["file_name"])
img1_path=os.path.join(self.root,pair["video_name"],pair["1"]["file_name"])
video_name=pair["video_name"]
img_name1=pair["0"]["file_name"]
img_name2=pair["1"]["file_name"]
cnt_0=pair["0"]["cnt"]
cnt_1=pair["1"]["cnt"]
pt_0=pair["0"]["pts"]
pt_1=pair["1"]["pts"]
if self.train:
pt_0=self.add_noise(pt_0)
pt_1=self.add_noise(pt_1)
bbox_0=pair["0"]["bboxes"]
bbox_1=pair["1"]["bboxes"]
id_0=pair["0"]["ids"]
id_1=pair["1"]["ids"]
if self.train and (pair["0"]["height"]!=pair["1"]["height"] or pair["0"]["width"]!=pair["1"]["width"] or cnt_0==0 or cnt_1==0):
print("error")
print(pair["0"]["height"],pair["1"]["height"],pair["0"]["width"],pair["1"]["width"],cnt_0,cnt_1)
return self.__getitem__((index+1)%len(self))
img0=self.transforms(image=np.array(Image.open(img0_path).convert("RGB")))["image"]
img1=self.transforms(image=np.array(Image.open(img1_path).convert("RGB")))["image"]
fused_pts_list0=[]
fused_pts_list1=[]
id_list0=[]
id_list1=[]
fused_num=0
fused_num1=0
id0_pt0_dict={id[0]:pt for id,pt in zip(id_0,pt_0)}
id1_pt1_dict={id[0]:pt for id,pt in zip(id_1,pt_1)}
independ0_list=[]
independ1_list=[]
for pt,id in zip(pt_0,id_0):
if id in id_1:
fused_pts_list0.append(pt)
id_list0.append(id)
fused_num+=1
pt1=id1_pt1_dict[id[0]]
fused_pts_list1.append(pt1)
id_list1.append(id)
else:
independ0_list.append(pt)
for pt,id in zip(pt_1,id_1):
if id not in id_0:
independ1_list.append(pt)
if len(independ0_list)==0:
independ0_list.append((0.25,0.25))
if len(independ1_list)==0:
independ1_list.append((0.75,0.75))
if self.train and (fused_num<2 or fused_num>30 or len(independ0_list)==0 or len(independ1_list)==0):
return self.__getitem__((index+1)%len(self))
independ_pts0=-1*np.ones((self.max_len,2))
independ_pts0[:len(independ0_list)]=np.array(independ0_list)
independ_pts1=-1*np.ones((self.max_len,2))
independ_pts1[:len(independ1_list)]=np.array(independ1_list)
x=torch.cat([img0,img1],dim=0)
fused_pts0=-1*np.ones((self.max_len,2))
fused_pts1=-1*np.ones((self.max_len,2))
if fused_num>0:
fused_pts0[:fused_num]=np.array(fused_pts_list0)
fused_pts1[:fused_num]=np.array(fused_pts_list1)
fused_pts1=torch.from_numpy(fused_pts1).float()
fused_pts0=torch.from_numpy(fused_pts0).float()
# cv2_img0=inverse_normalize(img0).permute(1,2,0).detach().cpu().numpy().astype(np.uint8)
# cv2_img1=inverse_normalize(img1).permute(1,2,0).detach().cpu().numpy().astype(np.uint8)
# img_pair=np.concatenate([cv2_img0,cv2_img1],axis=1)
# img_pair=cv2.cvtColor(img_pair,cv2.COLOR_RGB2BGR)
# for pt0,pt1 in zip(fused_pts0,fused_pts1):
# cv2.circle(img_pair,(int(pt0[0]*1280),int(pt0[1]*720)),5,(0,0,255),-1)
# cv2.circle(img_pair,(int(pt1[0]*1280)+1280,int(pt1[1]*720)),5,(0,0,255),-1)
# cv2.line(img_pair,(int(pt0[0]*1280),int(pt0[1]*720)),(int(pt1[0]*1280)+1280,int(pt1[1]*720)),(0,0,255),2)
# cv2.imwrite(f"outputs/vision/{video_name}_{img_name1}_{img_name2}.jpg",img_pair)
ref_pts=torch.stack([fused_pts0,fused_pts1],dim=0)
labels={
"h":pair["0"]["height"],
"w":pair["0"]["width"],
"gt_default_pts":pt_0,
"gt_duplicate_pts":pt_1,
"gt_fuse_pts0":fused_pts0,
"gt_fuse_pts1":fused_pts1,
"gt_default_num":pair["0"]["cnt"],
"gt_duplicate_num":pair["1"]["cnt"],
"gt_fuse_num":fused_num,
"video_name":video_name,
"img_name1":img_name1,
"img_name2":img_name2,
}
inputs={
"image_pair":x,
"ref_pts":ref_pts,
"ref_num":fused_num,
"independ_pts0":torch.from_numpy(independ_pts0).float(),
"independ_pts1":torch.from_numpy(independ_pts1).float(),
"independ_num0":len(independ0_list),
"independ_num1":len(independ1_list),
}
return inputs,labels
def build_dataset(root,annotation_dir,max_len,train=False,step=20,interval=1,force_last=False):
transforms=transform()
dataset=PairDataset(root,annotation_dir,max_len,transforms=transforms,train=train,step=step,interval=interval,force_last=force_last)
return dataset
def build_video_dataset(root,annotation_dir):
dataset=VideoDataset(root,annotation_dir,transforms=video_transform())
return dataset