if args.resume is not None:
print('Loading checkpoint / model...')
model_net.load_model(torch.load(args.resume)['model'])
else:
# Init weights of base
if not args.pretrained_base:
print('Init pretrained base...')
model_net.base_net.apply(init)
else:
print('Loading pretrained base...')
# Init the rest of weights
print('Init misc, extras, locations and confidences...')
model_net.apply_only_non_base(init)
"""
DATASET DEFINITION
"""
train_dataset = LogoDataset(root=os.path.dirname(os.path.abspath(__file__)) + '/dataset/logos/data', transform=SSD.Utils.Transform(AUGMENTERS, model_net.image_size))
val_dataset = LogoDataset(root=os.path.dirname(os.path.abspath(__file__)) + '/dataset/logos/data', transform=SSD.Utils.Transform(None, model_net.image_size))
size_train_dataset = int(0.8 * len(train_dataset))
indices = torch.randperm(len(train_dataset)).cpu()
train_dataset, val_dataset = torch.utils.data.dataset.Subset(train_dataset, indices[:size_train_dataset]), torch.utils.data.dataset.Subset(val_dataset, indices[size_train_dataset:])
"""
OPTIMIZER DEFINITION
"""
optimizer = torch.optim.Adam(net.parameters(), lr=args.lr, weight_decay=args.weight_decay)
