normalize,
])
criterion = HardNegativeContrastiveLoss().cuda()
coco_data_train = CocoCaptionsRV(args, sset="val", transform=prepro)
train_loader = DataLoader(coco_data_train,
batch_size=args.batch_size,
shuffle=False,
drop_last=True,
num_workers=args.workers,
collate_fn=collate_fn_padded,
pin_memory=True)
print("Initializing network...")
join_emb = joint_embedding(args).cuda()
# Grad false to all network
for param in join_emb.cap_emb.parameters():
param.requires_grad = False
for param in join_emb.img_emb.parameters():
param.requires_grad = False
optimizer = optim.Adam(filter(lambda p: p.requires_grad,
join_emb.parameters()),
lr=0.001)
eTime = time.time()
train_loss, batch_train, data_train = train(
train_loader,
help="The size of the batches",
type=int,
default=100)
parser.add_argument("-ct",
"--ctresh",
help="Thresholding coeeficient to binarize heat maps",
type=float,
default=0.45)
args = parser.parse_args()
print("Loading model from:", args.model_path)
checkpoint = torch.load(args.model_path,
map_location=lambda storage, loc: storage)
join_emb = joint_embedding(checkpoint['args_dict'])
join_emb.load_state_dict(checkpoint["state_dict"])
for param in join_emb.parameters():
param.requires_grad = False
join_emb = torch.nn.DataParallel(join_emb.cuda().eval())
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
in_dim = (400.0, 400.0)
prepro_val = transforms.Compose([
transforms.Resize((int(in_dim[0]), int(in_dim[1]))),
transforms.ToTensor(),
normalize,
transforms.RandomResizedCrop(256),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
normalize,
])
prepro_val = transforms.Compose(
[transforms.Resize((350, 350)),
transforms.ToTensor(), normalize])
end = time.time()
print("Initializing network ...", end=" ")
if args.resume: # Resume previous learning
checkpoint = torch.load(args.resume,
map_location=lambda storage, loc: storage)
join_emb = joint_embedding(checkpoint["args_dict"])
join_emb.load_state_dict(checkpoint["state_dict"])
join_emb = torch.nn.DataParallel(join_emb.cuda())
last_epoch = checkpoint["epoch"]
opti = checkpoint["optimizer"]
print("Load from epoch :", last_epoch)
last_epoch += 1
lr_scheduler = MultiStepLR(opti, args.lrd[1:], gamma=args.lrd[0])
lr_scheduler.step(last_epoch)
best_rec = checkpoint["best_rec"]
else:
# Create new model
