def main():
train_loader, val_loader = create_train_val_dataloaders(
args.train_data_fp, batch_size=args.batch_size)
model = DenseNet(drop_prob=0)
if use_cuda:
model = model.cuda()
train(model, train_loader, val_loader)
elif helper.params['dataset'] == 'dif':
num_classes = len(helper.labels)
else:
num_classes = 10
reseed(5)
if helper.params['model'] == 'densenet':
net = DenseNet(num_classes=num_classes,
depth=helper.params['densenet_depth'])
elif helper.params['model'] == 'resnet':
logger.info(f'Model size: {num_classes}')
net = models.resnet18(num_classes=num_classes)
elif helper.params['model'] == 'PretrainedRes':
net = models.resnet18(pretrained=True)
net.fc = nn.Linear(512, num_classes)
net = net.cuda()
elif helper.params['model'] == 'FlexiNet':
net = FlexiNet(3, num_classes)
elif helper.params['model'] == 'dif_inception':
net = inception_v3(pretrained=True, dif=True)
net.fc = nn.Linear(768, num_classes)
net.aux_logits = False
elif helper.params['model'] == 'inception':
net = inception_v3(pretrained=True)
net.fc = nn.Linear(2048, num_classes)
net.aux_logits = False
#model = torch.nn.DataParallel(model).cuda()
elif helper.params['model'] == 'mobilenet':
net = MobileNetV2(n_class=num_classes, input_size=64)
elif helper.params['model'] == 'word':
net = RNNModel(rnn_type='LSTM',