# Create dataloader
dataset = PanoDataset(root_dir=args.root_dir,
cat_list=[*args.input_cat, 'edge', 'cor'],
flip=False,
rotate=False,
gamma=False,
return_filenames=True)
# Prepare model
backend = args.backend.lower()
net = models[backend]()
net = nn.DataParallel(net).to(device)
net.load_state_dict(torch.load(args.ckpt))
# Start evaluation
test_losses = StatisticDict()
test_pano_losses = StatisticDict()
test_2d3d_losses = StatisticDict()
for ith, datas in enumerate(dataset):
print('processed %d batches out of %d' % (ith, len(dataset)),
end='\r',
flush=True)
x = torch.cat([datas[i] for i in range(len(args.input_cat))],
dim=0).numpy()
x_augmented, aug_type = augment(x, args.flip, args.rotate)
with torch.no_grad():
x_augmented = torch.FloatTensor(x_augmented).to(device)
de_list = net(x_augmented)
edg_tensor = torch.sigmoid(de_list[:, :-1])
def augment_undo(x_imgs_augmented, aug_type):
x_imgs = []
for x_img, aug in zip(x_imgs_augmented, aug_type):
if aug == 'flip':
x_imgs.append(np.flip(x_img, axis=-1))
elif aug.startswith('rotate'):
shift = int(aug.split()[-1])
x_imgs.append(np.roll(x_img, -shift, axis=-1))
elif aug == '':
x_imgs.append(x_img)
else:
raise NotImplementedError()
return np.array(x_imgs)
test_losses = StatisticDict()
for ith, datas in enumerate(dataset):
print('processed %d batches out of %d' % (ith, len(dataset)),
end='\r',
flush=True)
x = torch.cat([datas[i] for i in range(len(args.input_cat))],
dim=0).numpy()
x_augmented, aug_type = augment(x)
with torch.no_grad():
x_augmented = torch.FloatTensor(x_augmented).to(device)
en_list = encoder(x_augmented)
edg_de_list = edg_decoder(en_list[::-1])
cor_de_list = cor_decoder(en_list[-1:] + edg_de_list[:-1])
cor_tensor = torch.sigmoid(cor_de_list[-1])
num_workers=args.num_workers,
pin_memory=args.device.startswith('cuda'))
# Prepare model
encoder = Encoder(args.input_channels).to(device)
edg_decoder = Decoder(skip_num=2, out_planes=3).to(device)
cor_decoder = Decoder(skip_num=3, out_planes=1).to(device)
encoder.load_state_dict(torch.load('%s_encoder.pth' % args.path_prefix))
edg_decoder.load_state_dict(torch.load('%s_edg_decoder.pth' % args.path_prefix))
cor_decoder.load_state_dict(torch.load('%s_cor_decoder.pth' % args.path_prefix))
# Start evaluation
criti = nn.BCEWithLogitsLoss(reduction='none')
test_losses = StatisticDict()
for ith, datas in enumerate(loader):
print('processed %d batches out of %d' % (ith, len(loader)), end='\r', flush=True)
with torch.no_grad():
# Prepare data
x = torch.cat([datas[i]
for i in range(len(args.input_cat))], dim=1).to(device)
y_edg = datas[-2].to(device)
y_cor = datas[-1].to(device)
b_sz = x.size(0)
# Feedforward
en_list = encoder(x)
edg_de_list = edg_decoder(en_list[::-1])
cor_de_list = cor_decoder(en_list[-1:] + edg_de_list[:-1])
y_edg_ = edg_de_list[-1]
optimizer = optim.Adam([
*group_weight(encoder), *group_weight(edg_decoder),
*group_weight(cor_decoder)
],
lr=args.lr,
betas=(args.beta1, 0.999),
weight_decay=args.weight_decay)
else:
raise NotImplementedError()
# Init variable
args.warmup_iters = args.warmup_epochs * len(loader_train)
args.max_iters = args.epochs * len(loader_train)
args.running_lr = args.warmup_lr if args.warmup_epochs > 0 else args.lr
args.cur_iter = 0
train_losses = StatisticDict(winsz=100)
criti = nn.BCEWithLogitsLoss(reduction='none')
print(args)
print('%d iters per epoch for train' % len(loader_train))
print('%d iters per epoch for valid' % len(loader_valid))
print(' start training '.center(80, '='))
# Start training
for ith_epoch in range(1, args.epochs + 1):
for ith_batch, datas in enumerate(loader_train):
# Set learning rate
adjust_learning_rate(optimizer, args)
args.cur_iter += 1
# Prepare data
def augment_undo(x_imgs_augmented, aug_type):
x_imgs = []
for x_img, aug in zip(x_imgs_augmented, aug_type):
if aug == 'flip':
x_imgs.append(np.flip(x_img, axis=-1))
elif aug.startswith('rotate'):
shift = int(aug.split()[-1])
x_imgs.append(np.roll(x_img, -shift, axis=-1))
elif aug == '':
x_imgs.append(x_img)
else:
raise NotImplementedError()
return np.array(x_imgs)
test_losses = StatisticDict()
test_pano_losses = StatisticDict()
test_2d3d_losses = StatisticDict()
for ith, datas in enumerate(dataset):
print('processed %d batches out of %d' % (ith, len(dataset)),
end='\r',
flush=True)
x = torch.cat([datas[i] for i in range(len(args.input_cat))],
dim=0).numpy()
x_augmented, aug_type = augment(x)
with torch.no_grad():
x_augmented = torch.FloatTensor(x_augmented).to(device)
en_list = encoder(x_augmented)
edg_de_list = edg_decoder(en_list[::-1])