vgg16.load_state_dict(torch.load(f))
# device
use_gpu_if_available = True
use_CUDA = use_gpu_if_available and torch.cuda.is_available()
if use_CUDA:
device = torch.device('cuda')
else:
device = torch.device('cpu')
print('# using', device)
nn = vgg16.to(device)
nn.eval()
# dataset & dataloader
dT = ilsvrc2012.datasets('T')
bsize = 64
dl = torch.utils.data.DataLoader(dT,
batch_size=bsize,
shuffle=False,
pin_memory=use_CUDA,
num_workers=16)
nbatch = len(dl)
s1 = datetime.datetime.now()
ncorrect = 0
ntotal = 0
with torch.no_grad():
for ib, rv in enumerate(dl):
X, lab = rv[0].to(device), rv[1].to(device)
# loading VGG11-bn pretrained model
#vgg16 = torchvision.models.vgg16(pretrained=False)
vgg16 = DCTnet.DCTnet2v2()
print(vgg16)
# loading the parameters
fnParam_prev = f'data/ex20210103_trainDCT2v2_epoch{epoch_prev:03d}.pth'
if epoch_prev != 0:
with open(fnParam_prev, mode='rb') as f:
vgg16.load_state_dict(torch.load(f))
nn = vgg16.to(device)
nn.train()
# dataset & dataloader
dL = ilsvrc2012.datasets('L')
bsize = 256
dl = torch.utils.data.DataLoader(dL,
batch_size=bsize,
shuffle=True,
pin_memory=use_CUDA,
num_workers=16)
nbatch = len(dl)
# optimizer
#optimizer = optim.SGD(nn.parameters(), lr=0.01, momentum=0.9)
optimizer = optim.Adam(nn.parameters(), lr=0.0001, weight_decay=0.0)
nb = 1024 // bsize
ncList = np.zeros(nb, dtype=int)
lossList = np.zeros(nb)