}
#load image
source_image_root = os.path.join('/1116', 'SUN', 'rgb')
target_image_root = os.path.join('/1116', 'SUN', 'hha')
train_list = os.path.join('/1116', 'SUN', 'train_label.txt')
test_list = os.path.join('/1116', 'SUN', 'test_label.txt')
#phase = 'train' # train or test
data_list = {'train': train_list, 'test': test_list}
domain = 'target' #source or target
data_image_root = {'source': source_image_root, 'target': target_image_root}
dataset = {
phase: GetLoader(data_root=os.path.join(data_image_root[domain], phase),
data_list=data_list[phase],
transform=data_transforms[phase])
for phase in ['train', 'test']
}
dataset_sizes = {phase: len(dataset[phase]) for phase in ['train', 'test']}
dataloaders = {
phase: torch.utils.data.DataLoader(dataset=dataset[phase],
batch_size=batch_size,
shuffle=False,
num_workers=8)
for phase in ['train', 'test']
}
# Train and evaluate
def train_model(model, criterion, optimizer, scheduler, num_epochs=25):
else:
data_list = test_list
domain = 'source' #source or target
if domain == 'source':
data_image_root = source_image_root
feaname = 'rgb_' + phase + '_features.npy'
else:
data_image_root = target_image_root
feaname = 'hha_' + phase + '_features.npy'
model_name = 'sun_model_epoch_42.pth'
print(data_list)
dataset_source = GetLoader(data_root=os.path.join(data_image_root, phase),
data_list=data_list,
transform=data_transforms['test'])
dataloader_source = torch.utils.data.DataLoader(dataset=dataset_source,
batch_size=batch_size,
shuffle=False,
num_workers=8)
data_source_iter = iter(dataloader_source)
# load model
model_root = os.path.join('models', model_name)
my_net = torch.load(model_root)
for idx, m in enumerate(my_net.named_modules()):
print(idx, '-->', m)
# inter_feature = {}
# def make_hook(name, flag):