-
Notifications
You must be signed in to change notification settings - Fork 159
/
video_mAP.py
335 lines (273 loc) · 12.8 KB
/
video_mAP.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
import torch
import torch.nn as nn
import numpy as np
import os
import glob
from opts import parse_opts
from cfg import parse_cfg
from torch.utils.data import Dataset
from torchvision import transforms
from scipy.io import loadmat
from model import YOWO
from utils import *
from eval_results import *
opt = parse_opts()
dataset = opt.dataset
assert dataset == 'ucf101-24' or dataset == 'jhmdb-21', 'invalid dataset'
datacfg = opt.data_cfg
cfgfile = opt.cfg_file
gt_file = 'finalAnnots.mat' # Necessary for ucf
data_options = read_data_cfg(datacfg)
net_options = parse_cfg(cfgfile)[0]
loss_options = parse_cfg(cfgfile)[1]
base_path = data_options['base']
testlist = os.path.join(base_path, 'testlist_video.txt')
clip_duration = int(net_options['clip_duration'])
anchors = loss_options['anchors'].split(',')
anchors = [float(i) for i in anchors]
num_anchors = int(loss_options['num'])
num_classes = opt.n_classes
# Test parameters
conf_thresh = 0.005
nms_thresh = 0.4
eps = 1e-5
use_cuda = True
kwargs = {'num_workers': 0, 'pin_memory': True} if use_cuda else {}
# Create model
model = YOWO(opt)
model = model.cuda()
model = nn.DataParallel(model, device_ids=None) # in multi-gpu case
print(model)
# Load resume path
if opt.resume_path:
print("===================================================================")
print('loading checkpoint {}'.format(opt.resume_path))
checkpoint = torch.load(opt.resume_path)
model.load_state_dict(checkpoint['state_dict'])
model.eval()
print("===================================================================")
def get_clip(root, imgpath, train_dur, dataset):
im_split = imgpath.split('/')
num_parts = len(im_split)
class_name = im_split[-3]
file_name = im_split[-2]
im_ind = int(im_split[num_parts - 1][0:5])
if dataset == 'ucf101-24':
img_name = os.path.join(class_name, file_name, '{:05d}.jpg'.format(im_ind))
elif dataset == 'jhmdb-21':
img_name = os.path.join(class_name, file_name, '{:05d}.png'.format(im_ind))
labpath = os.path.join(base_path, 'labels', class_name, file_name, '{:05d}.txt'.format(im_ind))
img_folder = os.path.join(base_path, 'rgb-images', class_name, file_name)
max_num = len(os.listdir(img_folder))
clip = []
for i in reversed(range(train_dur)):
i_img = im_ind - i * 1
if i_img < 1:
i_img = 1
elif i_img > max_num:
i_img = max_num
if dataset == 'ucf101-24':
path_tmp = os.path.join(base_path, 'rgb-images', class_name, file_name, '{:05d}.jpg'.format(i_img))
elif dataset == 'jhmdb-21':
path_tmp = os.path.join(base_path, 'rgb-images', class_name, file_name, '{:05d}.png'.format(i_img))
clip.append(Image.open(path_tmp).convert('RGB'))
label = torch.zeros(50 * 5)
try:
tmp = torch.from_numpy(read_truths_args(labpath, 8.0 / clip[0].width).astype('float32'))
except Exception:
tmp = torch.zeros(1, 5)
tmp = tmp.view(-1)
tsz = tmp.numel()
if tsz > 50 * 5:
label = tmp[0:50 * 5]
elif tsz > 0:
label[0:tsz] = tmp
return clip, label, img_name
class testData(Dataset):
def __init__(self, root, shape=None, transform=None, clip_duration=16):
self.root = root
if dataset == 'ucf101-24':
self.label_paths = sorted(glob.glob(os.path.join(root, '*.jpg')))
elif dataset == 'jhmdb-21':
self.label_paths = sorted(glob.glob(os.path.join(root, '*.png')))
self.shape = shape
self.transform = transform
self.clip_duration = clip_duration
def __len__(self):
return len(self.label_paths)
def __getitem__(self, index):
assert index <= len(self), 'index range error'
label_path = self.label_paths[index]
clip, label, img_name = get_clip(self.root, label_path, self.clip_duration, dataset)
clip = [img.resize(self.shape) for img in clip]
if self.transform is not None:
clip = [self.transform(img) for img in clip]
clip = torch.cat(clip, 0).view((self.clip_duration, -1) + self.shape).permute(1, 0, 2, 3)
return clip, label, img_name
def video_mAP_ucf():
"""
Calculate video_mAP over the test dataset
"""
def truths_length(truths):
for i in range(50):
if truths[i][1] == 0:
return i
CLASSES = ('Basketball', 'BasketballDunk', 'Biking', 'CliffDiving', 'CricketBowling',
'Diving', 'Fencing', 'FloorGymnastics', 'GolfSwing', 'HorseRiding',
'IceDancing', 'LongJump', 'PoleVault', 'RopeClimbing', 'SalsaSpin',
'SkateBoarding', 'Skiing', 'Skijet', 'SoccerJuggling', 'Surfing',
'TennisSwing', 'TrampolineJumping', 'VolleyballSpiking', 'WalkingWithDog')
video_testlist = []
with open(testlist, 'r') as file:
lines = file.readlines()
for line in lines:
line = line.rstrip()
video_testlist.append(line)
detected_boxes = {}
gt_videos = {}
gt_data = loadmat(gt_file)['annot']
n_videos = gt_data.shape[1]
for i in range(n_videos):
video_name = gt_data[0][i][1][0]
if video_name in video_testlist:
n_tubes = len(gt_data[0][i][2][0])
v_annotation = {}
all_gt_boxes = []
for j in range(n_tubes):
gt_one_tube = []
tube_start_frame = gt_data[0][i][2][0][j][1][0][0]
tube_end_frame = gt_data[0][i][2][0][j][0][0][0]
tube_class = gt_data[0][i][2][0][j][2][0][0]
tube_data = gt_data[0][i][2][0][j][3]
tube_length = tube_end_frame - tube_start_frame + 1
for k in range(tube_length):
gt_boxes = []
gt_boxes.append(int(tube_start_frame+k))
gt_boxes.append(float(tube_data[k][0]))
gt_boxes.append(float(tube_data[k][1]))
gt_boxes.append(float(tube_data[k][0]) + float(tube_data[k][2]))
gt_boxes.append(float(tube_data[k][1]) + float(tube_data[k][3]))
gt_one_tube.append(gt_boxes)
all_gt_boxes.append(gt_one_tube)
v_annotation['gt_classes'] = tube_class
v_annotation['tubes'] = np.array(all_gt_boxes)
gt_videos[video_name] = v_annotation
for line in lines:
print(line)
line = line.rstrip()
test_loader = torch.utils.data.DataLoader(
testData(os.path.join(base_path, 'rgb-images', line),
shape=(224, 224), transform=transforms.Compose([
transforms.ToTensor()]), clip_duration=clip_duration),
batch_size=64, shuffle=False, **kwargs)
for batch_idx, (data, target, img_name) in enumerate(test_loader):
if use_cuda:
data = data.cuda()
with torch.no_grad():
data = Variable(data)
output = model(data).data
all_boxes = get_region_boxes_video(output, conf_thresh, num_classes, anchors, num_anchors, 0, 1)
for i in range(output.size(0)):
boxes = all_boxes[i]
boxes = nms(boxes, nms_thresh)
n_boxes = len(boxes)
# generate detected tubes for all classes
# save format: {img_name: {cls_ind: array[[x1,y1,x2,y2, cls_score], [], ...]}}
img_annotation = {}
for cls_idx in range(num_classes):
cls_idx += 1 # index begins from 1
cls_boxes = np.zeros([n_boxes, 5], dtype=np.float32)
for b in range(n_boxes):
cls_boxes[b][0] = max(float(boxes[b][0]-boxes[b][2]/2.0) * 320.0, 0.0)
cls_boxes[b][1] = max(float(boxes[b][1]-boxes[b][3]/2.0) * 240.0, 0.0)
cls_boxes[b][2] = min(float(boxes[b][0]+boxes[b][2]/2.0) * 320.0, 320.0)
cls_boxes[b][3] = min(float(boxes[b][1]+boxes[b][3]/2.0) * 240.0, 240.0)
cls_boxes[b][4] = float(boxes[b][5+(cls_idx-1)*2])
img_annotation[cls_idx] = cls_boxes
detected_boxes[img_name[i]] = img_annotation
iou_list = [0.05, 0.1, 0.2, 0.3, 0.5, 0.75]
for iou_th in iou_list:
print('iou is: ', iou_th)
print(evaluate_videoAP(gt_videos, detected_boxes, CLASSES, iou_th, True))
def video_mAP_jhmdb():
"""
Calculate video_mAP over the test set
"""
def truths_length(truths):
for i in range(50):
if truths[i][1] == 0:
return i
CLASSES = ('brush_hair', 'catch', 'clap', 'climb_stairs', 'golf',
'jump', 'kick_ball', 'pick', 'pour', 'pullup', 'push',
'run', 'shoot_ball', 'shoot_bow', 'shoot_gun', 'sit',
'stand', 'swing_baseball', 'throw', 'walk', 'wave')
with open(testlist, 'r') as file:
lines = file.readlines()
detected_boxes = {}
gt_videos = {}
for line in lines:
print(line)
line = line.rstrip()
test_loader = torch.utils.data.DataLoader(
testData(os.path.join(base_path, 'rgb-images', line),
shape=(224, 224), transform=transforms.Compose([
transforms.ToTensor()]), clip_duration=clip_duration),
batch_size=1, shuffle=False, **kwargs)
video_name = ''
v_annotation = {}
all_gt_boxes = []
t_label = -1
for batch_idx, (data, target, img_name) in enumerate(test_loader):
path_split = img_name[0].split('/')
if video_name == '':
video_name = os.path.join(path_split[0], path_split[1])
if use_cuda:
data = data.cuda()
with torch.no_grad():
data = Variable(data)
output = model(data).data
all_boxes = get_region_boxes_video(output, conf_thresh, num_classes, anchors, num_anchors, 0, 1)
for i in range(output.size(0)):
boxes = all_boxes[i]
boxes = nms(boxes, nms_thresh)
n_boxes = len(boxes)
truths = target[i].view(-1, 5)
num_gts = truths_length(truths)
if t_label == -1:
t_label = int(truths[0][0]) + 1
# generate detected tubes for all classes
# save format: {img_name: {cls_ind: array[[x1,y1,x2,y2, cls_score], [], ...]}}
img_annotation = {}
for cls_idx in range(num_classes):
cls_idx += 1 # index begins from 1
cls_boxes = np.zeros([n_boxes, 5], dtype=np.float32)
for b in range(n_boxes):
cls_boxes[b][0] = max(float(boxes[b][0]-boxes[b][2]/2.0) * 320.0, 0.0)
cls_boxes[b][1] = max(float(boxes[b][1]-boxes[b][3]/2.0) * 240.0, 0.0)
cls_boxes[b][2] = min(float(boxes[b][0]+boxes[b][2]/2.0) * 320.0, 320.0)
cls_boxes[b][3] = min(float(boxes[b][1]+boxes[b][3]/2.0) * 240.0, 240.0)
cls_boxes[b][4] = float(boxes[b][5+(cls_idx-1)*2])
img_annotation[cls_idx] = cls_boxes
detected_boxes[img_name[0]] = img_annotation
# generate corresponding gts
# save format: {v_name: {tubes: [[frame_index, x1,y1,x2,y2]], gt_classes: vlabel}}
gt_boxes = []
for g in range(num_gts):
gt_boxes.append(int(path_split[2][:5]))
gt_boxes.append(float(truths[g][1]-truths[g][3]/2.0) * 320.0)
gt_boxes.append(float(truths[g][2]-truths[g][4]/2.0) * 240.0)
gt_boxes.append(float(truths[g][1]+truths[g][3]/2.0) * 320.0)
gt_boxes.append(float(truths[g][2]+truths[g][4]/2.0) * 240.0)
all_gt_boxes.append(gt_boxes)
v_annotation['gt_classes'] = t_label
v_annotation['tubes'] = np.expand_dims(np.array(all_gt_boxes), axis=0)
gt_videos[video_name] = v_annotation
iou_list = [0.05, 0.1, 0.2, 0.3, 0.5, 0.75]
for iou_th in iou_list:
print('iou is: ', iou_th)
print(evaluate_videoAP(gt_videos, detected_boxes, CLASSES, iou_th, True))
if __name__ == '__main__':
if opt.dataset == 'ucf101-24':
video_mAP_ucf()
elif opt.dataset == 'jhmdb-21':
video_mAP_jhmdb()