-
Notifications
You must be signed in to change notification settings - Fork 1
/
load_ucf101_pytorch_gaus.py
338 lines (274 loc) · 13.6 KB
/
load_ucf101_pytorch_gaus.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
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
import os
import time
import numpy as np
import random
from threading import Thread
from scipy.io import loadmat
from skvideo.io import vread
import pdb
import torch
from torch.utils.data import Dataset
import pickle
import cv2
from scipy.stats import norm
'''
Loads in videos for the 24 class subset of UCF-101.
The data is assumed to be organized in a folder (dataset_dir):
-Subfolder UCF101_vids contains the videos
-Subfolder UCF101_Annotations contains the .mat annotation files
UCF101DataLoader loads in the videos and formats their annotations on seperate threads.
-Argument train_or_test determines if you want to load in training or testing data
-Argument sec_to_wait determines the amount of time to wait to load in data
-Argument n_threads determines the number of threads which will be created to load data
Calling .get_video() returns (video, bboxes, label)
-The videos are in the shape (F, H, W, 3).
-The bounding boxes are loaded in as heat maps of size (F, H, W, 1) where 1 is forground and 0 is background.
-The label is an integer corresponding to the action class.
'''
class UCF101DataLoader(Dataset):
'Prunes UCF101-24 data'
def __init__(self, name, clip_shape, batch_size, percent='5', use_random_start_frame=False):
self._dataset_dir = './data/UCF101' # Path to UCF101
if name == 'train':
self.vid_files = self.get_det_annots_prepared(percent=percent)
self.shuffle = True
self.name = 'train'
else:
self.vid_files = self.get_det_annots_test_prepared()
self.shuffle = False
self.name = 'test'
self._use_random_start_frame = use_random_start_frame
self._height = clip_shape[0]
self._width = clip_shape[1]
#self._channels = channels
self._batch_size = batch_size
self._size = len(self.vid_files)
self.indexes = np.arange(self._size)
def get_det_annots_prepared(self, percent='5'):
import pickle
training_annot_file = 'training_annots_multi.pkl'
#training_annot_file = 'training_annots_multi_'+percent+'perRand.pkl'
with open(training_annot_file, 'rb') as tr_rid:
training_annotations = pickle.load(tr_rid)
print("Training samples from :", training_annot_file)
training_annotations.extend(training_annotations)
return training_annotations
def get_det_annots_test_prepared(self):
import pickle
with open('testing_annots.pkl', 'rb') as ts_rid:
testing_annotations = pickle.load(ts_rid)
return testing_annotations
def __len__(self):
'Denotes the number of batches per epoch'
return len(self.vid_files)
def __getitem__(self, index):
depth = 8
video_rgb = np.zeros((depth, self._height, self._width, 3))
label_cls = np.zeros((depth, self._height, self._width, 1)) # FG/BG or actor (person only) for this dataset
mask_cls = np.zeros((depth, self._height, self._width, 1))
v_name, anns = self.vid_files[index]
clip, bbox_clip, label, annot_frames = self.load_video(v_name, anns)
if clip is None:
video_rgb = np.transpose(video_rgb, [3, 0, 1, 2]) #moving channels to first position
video_rgb = torch.from_numpy(video_rgb)
label_cls = np.transpose(label_cls, [3, 0, 1, 2])
label_cls = torch.from_numpy(label_cls)
mask_cls = np.transpose(mask_cls, [3, 0, 1, 2])
mask_cls = torch.from_numpy(mask_cls)
sample = {'data':video_rgb,'segmentation':label_cls,'action':torch.Tensor([0]),'mask_cls':mask_cls}
return sample
vlen, clip_h, clip_w, _ = clip.shape
vskip = 2
if len(annot_frames) == 1:
selected_annot_frame = annot_frames[0]
else:
if len(annot_frames) <= 0:
print('annot index error for', v_name, ', ', len(annot_frames), ', ', annot_frames)
video_rgb = np.transpose(video_rgb, [3, 0, 1, 2]) #moving channels to first position
video_rgb = torch.from_numpy(video_rgb)
label_cls = np.transpose(label_cls, [3, 0, 1, 2])
label_cls = torch.from_numpy(label_cls)
mask_cls = np.transpose(mask_cls, [3, 0, 1, 2])
mask_cls = torch.from_numpy(mask_cls)
sample = {'data':video_rgb,'segmentation':label_cls,'action':torch.Tensor([0]),'mask_cls':mask_cls}
return sample
annot_idx = np.random.randint(0,len(annot_frames))
selected_annot_frame = annot_frames[annot_idx]
start_frame = selected_annot_frame - int((depth * vskip)/2)
if start_frame < 0:
vskip = 1
start_frame = selected_annot_frame - int((depth * vskip)/2)
if start_frame < 0:
start_frame = 0
vskip = 1
if selected_annot_frame >= vlen:
video_rgb = np.transpose(video_rgb, [3, 0, 1, 2]) #moving channels to first position
video_rgb = torch.from_numpy(video_rgb)
label_cls = np.transpose(label_cls, [3, 0, 1, 2])
label_cls = torch.from_numpy(label_cls)
mask_cls = np.transpose(mask_cls, [3, 0, 1, 2])
mask_cls = torch.from_numpy(mask_cls)
sample = {'data':video_rgb,'segmentation':label_cls,'action':torch.Tensor([0]),'mask_cls':mask_cls}
return sample
if start_frame + (depth * vskip) >= vlen:
start_frame = vlen - (depth * vskip)
#Random start frame
if self._use_random_start_frame:
random_start_frame_btm = selected_annot_frame - (depth * vskip) + 2
if random_start_frame_btm < 0:
random_start_frame_btm = 0
random_start_frame_top = selected_annot_frame - 2
if random_start_frame_top <= random_start_frame_btm:
random_start_frame = start_frame
else:
random_start_frame = np.random.randint(random_start_frame_btm, random_start_frame_top)
if random_start_frame + (depth * vskip) >= vlen:
random_start_frame = vlen - (depth * vskip)
start_frame = random_start_frame
span = (np.arange(depth)*vskip)
span += start_frame
video = clip[span]
bbox_clip = bbox_clip[span]
closest_fidx = np.argmin(np.abs(span-selected_annot_frame))
sigma = 4/3. # 2.5
gaus_vals = norm.pdf(np.arange(-4,5), 0, sigma)
gaus_vals = gaus_vals / np.max(gaus_vals)
gaus_mid = 5
if self.name == 'train':
#start_pos = np.random.randint(0,16)
start_pos_h = np.random.randint(0,clip_h - 224) #self._height)
start_pos_w = np.random.randint(0,clip_w - 296) #self._width)
else:
# center crop for validation
start_pos_h = int((clip_h - 224) / 2)
start_pos_w = int((clip_w - 296) / 2)
final_gaus_mask = np.zeros((depth))
for j in range(video.shape[0]):
img = video[j]
img = img[start_pos_h:start_pos_h+224, start_pos_w:start_pos_w+296,:]
img = cv2.resize(img, (self._height,self._width), interpolation=cv2.INTER_LINEAR)
img = img / 255.
video_rgb[j] = img
valid_frame = False
if vskip == 2:
if span[j] in annot_frames or span[j]+1 in annot_frames:
valid_frame = True
elif vskip == 1:
if span[j] in annot_frames:
valid_frame = True
bbox_img = bbox_clip[j]
bbox_img = bbox_img[start_pos_h:start_pos_h+224, start_pos_w:start_pos_w+296,:]
bbox_img = cv2.resize(bbox_img, (self._height,self._width), interpolation=cv2.INTER_LINEAR)
label_cls[j, bbox_img > 0, 0] = 1.
if valid_frame:
gaus_range = [np.max((0,gaus_mid-j-1)), np.min((9, gaus_mid+depth-j-1))]
frame_range = [np.max((0, j-gaus_vals.shape[0]//2)), np.min((8, j+gaus_vals.shape[0]//2+1))]
mask_vals = gaus_vals[gaus_range[0]:gaus_range[1]]
# Max
new_mask_vals = final_gaus_mask[frame_range[0]:frame_range[1]]
new_mask_vals[mask_vals>new_mask_vals] = (0*new_mask_vals[mask_vals>new_mask_vals]) + mask_vals[mask_vals>new_mask_vals]
final_gaus_mask[frame_range[0]:frame_range[1]] = np.copy(new_mask_vals)
# Max
#print(final_gaus_mask)
final_gaus_mask = final_gaus_mask.reshape((final_gaus_mask.shape[0], 1, 1))
mask_cls[:, :, :, 0] = final_gaus_mask
mask_cls[mask_cls>1.] = 1.
#if np.random.randint(0,2) == 1:
if index<(self._size/2): # If first half, horizontal flip. Assumes data duplicated beforehand in load
video_rgb[:,:,:,:] = video_rgb[:,:,::-1,:]
label_cls[:,:,:,:] = label_cls[:,:,::-1,:]
mask_cls[:,:,:,:] = mask_cls[:,:,::-1,:]
video_rgb = np.transpose(video_rgb, [3, 0, 1, 2]) #moving channels to first position
video_rgb = torch.from_numpy(video_rgb)
label_cls = np.transpose(label_cls, [3, 0, 1, 2])
label_cls = torch.from_numpy(label_cls)
mask_cls = np.transpose(mask_cls, [3, 0, 1, 2])
mask_cls = torch.from_numpy(mask_cls)
action_tensor = torch.Tensor([label])
sample = {'data':video_rgb,'segmentation':label_cls,'action':action_tensor, 'mask_cls': mask_cls}
return sample
def load_video(self, video_name, annotations):
video_dir = os.path.join(self._dataset_dir, 'UCF101_Videos/%s.avi' % video_name)
try:
video = vread(str(video_dir)) # Reads in the video into shape (F, H, W, 3)
except:
print('Error:', str(video_dir))
return None, None, None, None
# creates the bounding box annotation at each frame
n_frames, h, w, ch = video.shape
bbox = np.zeros((n_frames, h, w, 1), dtype=np.uint8)
label = -1
#multi frame mode
annot_idx = 0
if len(annotations) > 1:
annot_idx = np.random.randint(0,len(annotations))
if self.name == 'train':
multi_frame_annot = annotations[annot_idx][4]
else:
multi_frame_annot = []
bbox_annot = np.zeros((n_frames, h, w, 1), dtype=np.uint8)
for ann in annotations:
if not self.name == 'train':
multi_frame_annot.extend(ann[4])
start_frame, end_frame, label = ann[0], ann[1], ann[2]
collect_annots = []
for f in range(start_frame, min(n_frames, end_frame+1)):
try:
x, y, w, h = ann[3][f-start_frame]
bbox[f, y:y+h, x:x+w, :] = 1
if f in ann[4]:
collect_annots.append([x,y,w,h])
except:
print('ERROR LOADING ANNOTATIONS')
print(start_frame, end_frame)
print(video_dir)
exit()
select_annots = ann[4]
select_annots.sort()
if len(collect_annots) == 0:
continue
[x, y, w, h] = collect_annots[0]
if len(collect_annots) == 1:
bbox_annot[start_frame:end_frame, y:y+h, x:x+w, :] = 1
else:
bbox_annot[start_frame:select_annots[0], y:y+h, x:x+w, :] = 1
for i in range(len(collect_annots)-1):
frame_diff = select_annots[i+1] - select_annots[i]
if frame_diff > 1:
[x, y, w, h] = collect_annots[i]
pt1 = np.array([x, y, x+w, y+h])
[x, y, w, h] = collect_annots[i+1]
pt2 = np.array([x, y, x+w, y+h])
points = np.linspace(pt1, pt2, frame_diff).astype(np.int32)
for j in range(points.shape[0]):
[x1, y1, x2, y2] = points[j]
bbox_annot[select_annots[i]+j, y1:y2, x1:x2, :] = 1
else:
[x, y, w, h] = collect_annots[i]
bbox_annot[select_annots[i], y:y+h, x:x+w, :] = 1
[x, y, w, h] = collect_annots[-1]
bbox_annot[select_annots[-1]:end_frame, y:y+h, x:x+w, :] = 1
multi_frame_annot = list(set(multi_frame_annot))
if self.name == 'train':
return video, bbox_annot, label, multi_frame_annot
else:
return video, bbox, label, multi_frame_annot
if __name__ == '__main__':
import imageio
name='train'
clip_shape=[224,224]
channels=3
batch_size = 5
dataloader = UCF101DataLoader(name, clip_shape, batch_size, '5', False)
print(len(dataloader))
#exit()
index = 0
while index<2:
d1 = dataloader.__getitem__(index)
print(frm_idx)
with imageio.get_writer('./results/{:02d}_gt.gif'.format(index), mode='I') as writer:
for i in range(clip.shape[1]):
image = (clip[0,i]*255).astype(np.uint8)
writer.append_data(image)
print("Done for ", index)
index += 1