def main():
global best_loss
start_epoch = 0 # start from epoch 0 or last checkpoint epoch
val_loader = torch.utils.data.DataLoader(
davis.DavisSet(params, is_train=False),
batch_size=int(params['batchSize']), shuffle=False,
num_workers=args.workers, pin_memory=True)
model = tc.TimeCycle()
model = Wrap(model, 'forward_affinity')
model = torch.nn.DataParallel(model).cuda()
cudnn.benchmark = False
print(' Total params: %.2fM' % (sum(p.numel() for p in model.parameters())/1000000.0))
# Load checkpoint.
if os.path.isfile(args.resume):
print('==> Resuming from checkpoint..')
checkpoint = torch.load(args.resume)
partial_load(checkpoint['state_dict'], model)
del checkpoint
model.eval()
if not os.path.exists(args.save_path):
os.makedirs(args.save_path)
print('\Testing')
test_loss = test(val_loader, model, 1, use_cuda)
def main():
global best_loss
start_epoch = 0 # start from epoch 0 or last checkpoint epoch
args.kldv_coef = 1
args.long_coef = 1
args.frame_transforms = 'crop'
args.frame_aug = 'grid'
args.npatch = 49
args.img_size = 256
args.pstride = [0.5, 0.5]
args.patch_size = [64, 64, 3]
args.visualize = False
model = tc.TimeCycle(args, vis=vis).cuda()
params['mapScale'] = model(torch.zeros(1, 10, 3, 320, 320).cuda(),
just_feats=True)[1].shape[-2:]
params['mapScale'] = 320 // np.array(params['mapScale'])
val_loader = torch.utils.data.DataLoader(
davis.DavisSet(params, is_train=False) if not 'jhmdb' in args.filelist else \
jhmdb.JhmdbSet(params, is_train=False),
batch_size=int(params['batchSize']), shuffle=False,
num_workers=args.workers, pin_memory=True)
cudnn.benchmark = False
print(' Total params: %.2fM' %
(sum(p.numel() for p in model.parameters()) / 1000000.0))
# Load checkpoint.
if os.path.isfile(args.resume):
print('==> Resuming from checkpoint..')
checkpoint = torch.load(args.resume)
utils.partial_load(checkpoint['model'], model, skip_keys=['head'])
del checkpoint
model.eval()
# model = torch.nn.DataParallel(model).cuda() # model = model.cuda()
model = model.cuda()
if not os.path.exists(args.save_path):
os.makedirs(args.save_path)
print('\Testing')
# with torch.no_grad():
test_loss = test(val_loader, model, 1, use_cuda, args)
def main():
global best_loss
start_epoch = 0 # start from epoch 0 or last checkpoint epoch
model = tc.TimeCycle(args).cuda()
model = Wrap(model)
params['mapScale'] = model(torch.zeros(1, 10, 3, 320, 320).cuda(),
None,
True,
func='forward')[1].shape[-2:]
params['mapScale'] = 320 // np.array(params['mapScale'])
val_loader = torch.utils.data.DataLoader(davis.DavisSet(params,
is_train=False),
batch_size=int(
params['batchSize']),
shuffle=False,
num_workers=args.workers,
pin_memory=True)
cudnn.benchmark = False
print(' Total params: %.2fM' %
(sum(p.numel() for p in model.parameters()) / 1000000.0))
# Load checkpoint.
if os.path.isfile(args.resume):
print('==> Resuming from checkpoint..')
checkpoint = torch.load(args.resume)
# model.model.load_state_dict(checkpoint['model'])
utils.partial_load(checkpoint['model'], model.model)
del checkpoint
model.eval()
model = torch.nn.DataParallel(model).cuda() # model = model.cuda()
if not os.path.exists(args.save_path):
os.makedirs(args.save_path)
print('\Testing')
with torch.no_grad():
test_loss = test(val_loader, model, 1, use_cuda)