model = None
if args['load'] :
model = torch.load(StoragePath + args['load'])
else :
model = ImageNet().half().cuda()
PruneList = ['Conv2d', 'Linear']
PrunePair = ()
for m in model.modules():
if len(list(m.children())) == 0 and m.__class__.__name__ in PruneList: # skip for non-leaf module
PrunePair += ((m, 'weight'), )
prune.RandomStructured.apply(module = m, name = 'weight', amount = 0.5)
# prune.global_unstructured(PrunePair, pruning_method = prune.RandomStructured, amount = 0.6)
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(model.parameters(), lr = args['learning_rate'], momentum = 0.9)
print('Training :')
lastaccuracy = 0
# model.eval()
# with torch.no_grad() :
# lastaccuracy = forward('Validation', validationloader, model)
if args['save'] :
torch.save(model, StoragePath + args['save'])
for epoch in range(args['epoch']): # loop over the dataset multiple times
print('\n\tEpoch : ' + str(epoch))
loss_print = 0
MODEL_UPDATE_ITER = 0
# get loader
train_loader = get_train_loader(opt=opt)
# define net
imageNet = ImageNet()
imageNet.cuda()
# text net
tokenizer = BertTokenizer.from_pretrained(
'/home/poac/code/Multi_modal_Retrieval/experiments/pretrained_models/bert-base-uncased-vocab.txt'
)
textNet = TextNet(code_length=opt.hashbits)
textNet.cuda()
# embedding net
embNet = EmbNet(opt)
embNet.cuda()
optimizer = optim.Adam(
list(imageNet.parameters()) + list(textNet.parameters()),
lr=opt.lr,
weight_decay=opt.weight_decay) #+list(textExtractor.parameters())
# train and test on 10 epoch
for epoch in tqdm(range(start_epoch, start_epoch + opt.max_epochs + 1)):
train(opt, epoch)
if epoch % 1 == 0:
test(opt, epoch)
shuffle=False,
num_workers=args['thread'])
testloader = torch.utils.data.DataLoader(testset,
batch_size=args['batch_size'],
shuffle=False,
num_workers=args['thread'])
model = None
if args['load']:
model = torch.load(args['load'])
else:
model = ImageNet().cuda()
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(model.parameters(),
lr=args['learning_rate'],
momentum=0.9)
print('Training :')
avgloss = 0.0
avgcorrect = [0.0] * 11
cases = [0.0] * 11
lastaccuracy = 0
model.eval()
with torch.no_grad():
for i, (inputs, labels) in enumerate(validationloader):
inputs = inputs.cuda()
labels = labels.cuda()