parser.add_argument(
'--whitenm',
type=float,
default=1.0,
help='whitening multiplier, default is 1.0 (i.e. no multiplication)')
args = parser.parse_args()
args.iscuda = common.torch_set_gpu(args.gpu)
if args.aqe is not None:
args.aqe = {'k': args.aqe[0], 'alpha': args.aqe[1]}
if args.adba is not None:
args.adba = {'k': args.adba[0], 'alpha': args.adba[1]}
dl.download_dataset(args.dataset)
dataset = datasets.create(args.dataset)
print("Test dataset:", dataset)
net = load_model(args.checkpoint, args.iscuda)
if args.whiten:
net.pca = net.pca[args.whiten]
args.whiten = {
'whitenp': args.whitenp,
'whitenv': args.whitenv,
'whitenm': args.whitenm
}
else:
net.pca = None
args.whiten = None
parser.add_argument('--batch_size', type=int, default=320, help='size of batch imags')
parser.add_argument('--epochs', type=int, default=300, help='train epochs')
parser.add_argument('--saved', type=str, default='./experiments/ICIAR+no_pretrained/', help='train epochs')
args = parser.parse_args()
gpu_ids = args.gpu[0].split(',')
gpu_ids = [int(i) for i in gpu_ids]
args.iscuda = common.torch_set_gpu(gpu_ids)
if args.aqe is not None:
args.aqe = {'k': args.aqe[0], 'alpha': args.aqe[1]}
if args.adba is not None:
args.adba = {'k': args.adba[0], 'alpha': args.adba[1]}
#load dataset
dataset = datasets.create(args.dataset)
test_dataset = datasets.create(args.test_dataset)
#load model
print("With %s Train Model:" %(args.dataset))
net, start_epoch = load_model(args.checkpoint, args.iscuda)
net.cuda()
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
net = torch.nn.DataParallel(net,device_ids = [0])
net.to(device)
#define optimizer parameter
net.pca = net.module.pca
net.preprocess = net.module.preprocess
net.iscuda = net.module.iscuda
criterion = APLoss()
help='number of thread workders')
parser.add_argument('--gpu',
type=int,
default=0,
nargs='+',
help='GPU ids')
parser.add_argument('--cache',
type=str,
required=True,
help='path to cache files')
args = parser.parse_args()
assert args.buffer_size % args.batch_size == 0
args.iscuda = common.torch_set_gpu(args.gpu)
train_set = datasets.create('Landmarks_clean')
val_set = datasets.create('RParis6K')
model_options = {
'arch': args.arch,
'out_dim': args.out_dim,
'pooling': args.pooling,
'gemp': args.gemp
}
start_epoch = 0
if os.path.isfile(args.resume):
checkpoint = common.load_checkpoint(args.resume, args.iscuda)
net = nets.create_model(pretrained='', **model_options)
net = common.switch_model_to_cuda(net, args.iscuda, checkpoint)
net.load_state_dict(checkpoint['state_dict'])
parser.add_argument('--aqe', type=int, nargs='+', help='alpha-query expansion paramenters')
parser.add_argument('--adba', type=int, nargs='+', help='alpha-database augmentation paramenters')
parser.add_argument('--whitenp', type=float, default=0.25, help='whitening power, default is 0.5 (i.e., the sqrt)')
parser.add_argument('--whitenv', type=int, default=None, help='number of components, default is None (i.e. all components)')
parser.add_argument('--whitenm', type=float, default=1.0, help='whitening multiplier, default is 1.0 (i.e. no multiplication)')
args = parser.parse_args()
args.iscuda = common.torch_set_gpu(args.gpu)
if args.aqe is not None: args.aqe = {'k': args.aqe[0], 'alpha': args.aqe[1]}
if args.adba is not None: args.adba = {'k': args.adba[0], 'alpha': args.adba[1]}
dl.download_dataset(args.dataset)
dataset = datasets.create(args.dataset)
print("Test dataset:", dataset)
net = load_model(args.checkpoint, args.iscuda)
if args.center_bias:
assert hasattr(net,'center_bias')
net.center_bias = args.center_bias
if hasattr(net, 'module') and hasattr(net.module,'center_bias'):
net.module.center_bias = args.center_bias
if args.whiten and not hasattr(net, 'pca'):
# Learn PCA if necessary
if os.path.exists(args.whiten):
with open(args.whiten, 'rb') as f:
net.pca = pkl.load(f)