train_augmentation = net.tsn.get_augmentation()
input_mean = net.tsn.input_mean
input_std = net.tsn.input_std
if modality != 'RGBDiff':
normalize = GroupNormalize(input_mean, input_std)
else:
normalize = IdentityTransform()
train_loader = torch.utils.data.DataLoader(
TSNDataSet("", test_list, num_segments=num_segments,
new_length=data_length,
modality=modality,
image_tmpl="img_{:05d}.jpg" if modality in ["RGB", "RGBDiff"] else flow_prefix+"{}_{:05d}.jpg",
test_mode=True,
transform=torchvision.transforms.Compose([
GroupCenterCrop([224, 224]),
Stack(roll=arch == 'BNInception'),
ToTorchFormatTensor(div=arch != 'BNInception'),
normalize,
])
),
batch_size=batch_size, shuffle=False,
num_workers=workers, pin_memory=True, drop_last=False)
print("Length of dataset is {}".format(len(train_loader)))
''' Start Testing Process '''
accur = []
gt = []
for epoch in range(1):
for idx, (input, target, indice) in enumerate(train_loader):
train_augmentation = net.tsn.get_augmentation()
input_mean = net.tsn.input_mean
input_std = net.tsn.input_std
if modality != 'RGBDiff':
normalize = GroupNormalize(input_mean, input_std)
else:
normalize = IdentityTransform()
train_loader = torch.utils.data.DataLoader(TSNDataSet(
"",
train_list,
num_segments=num_segments,
new_length=data_length,
modality=modality,
image_tmpl="img_{:05d}.jpg"
if modality in ["RGB", "RGBDiff"] else flow_prefix + "{}_{:05d}.jpg",
transform=torchvision.transforms.Compose([
train_augmentation,
Stack(roll=arch == 'BNInception'),
ToTorchFormatTensor(div=arch != 'BNInception'),
normalize,
])),
batch_size=batch_size,
shuffle=True,
num_workers=workers,
pin_memory=True,
drop_last=True)
print("Length of dataset is {}".format(len(train_loader)))
''' Start Training Process '''
for epoch in range(400):
for idx, (input, target, indice) in enumerate(train_loader):