def get_training_set(opt, spatial_transform, temporal_transform,
target_transform):
assert opt.dataset in ['kinetics', 'activitynet', 'ucf101', 'hmdb51']
if opt.dataset == 'kinetics':
training_data = Kinetics(opt.video_path,
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'activitynet':
training_data = ActivityNet(opt.video_path,
opt.annotation_path,
'training',
False,
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'ucf101':
training_data = UCF101(opt.video_path,
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
elif opt.dataset == 'hmdb51':
training_data = HMDB51(opt.video_path,
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
return training_data
def get_test_set(opt, spatial_transform, temporal_transform, target_transform):
assert opt.dataset in ['kinetics', 'activitynet', 'ucf101', 'hmdb51']
assert opt.test_subset in ['val', 'test']
if opt.test_subset == 'val':
subset = 'validation'
elif opt.test_subset == 'test':
subset = 'testing'
if opt.dataset == 'kinetics':
test_data = Kinetics(opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration,
sample_stride=opt.sample_stride)
elif opt.dataset == 'activitynet':
test_data = ActivityNet(opt.video_path,
opt.annotation_path,
subset,
True,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'ucf101':
test_data = UCF101(opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'hmdb51':
test_data = HMDB51(opt.video_path,
opt.annotation_path,
subset,
0,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
return test_data
def get_validation_set(opt, spatial_transform, temporal_transform,
target_transform):
assert opt.dataset in ['kinetics', 'activitynet', 'ucf101', 'hmdb51']
if opt.dataset == 'kinetics':
validation_data = Kinetics(opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'activitynet':
validation_data = ActivityNet(opt.video_path,
opt.annotation_path,
'validation',
False,
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'ucf101':
validation_data = UCF101(opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
elif opt.dataset == 'hmdb51':
validation_data = HMDB51(opt.video_path,
opt.annotation_path,
'validation',
opt.n_val_samples,
spatial_transform,
temporal_transform,
target_transform,
sample_duration=opt.sample_duration)
return validation_data
if not opt.no_cuda:
criterion = criterion.cuda()
if not opt.no_train:
spatial_transform = Compose([
MultiScaleCornerCrop(opt.scales, opt.sample_size),
RandomHorizontalFlip(),
ToTensor(opt.norm_value),
Normalize(opt.mean, [1, 1, 1])
])
temporal_transform = TemporalRandomCrop(opt.sample_duration)
target_transform = ClassLabel()
if opt.dataset == 'kinetics':
training_data = Kinetics(opt.video_path,
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
else:
training_data = ActivityNet(opt.video_path,
opt.annotation_path,
'training',
spatial_transform=spatial_transform,
temporal_transform=temporal_transform,
target_transform=target_transform)
train_loader = torch.utils.data.DataLoader(training_data,
batch_size=opt.batch_size,
shuffle=True,
num_workers=opt.n_threads,
pin_memory=True)
train_logger = Logger(os.path.join(opt.result_path, 'train.log'),