def load_hyper(args):
data, labels, numclass = auxil.loadData(args.dataset,
num_components=args.components)
pixels, labels = auxil.createImageCubes(data,
labels,
windowSize=args.spatialsize,
removeZeroLabels=True)
bands = pixels.shape[-1]
numberofclass = len(np.unique(labels))
x_train, x_test, y_train, y_test = auxil.split_data(
pixels, labels, args.tr_percent)
del pixels, labels
train_hyper = HyperData(
(np.transpose(x_train, (0, 3, 1, 2)).astype("float32"), y_train), None)
test_hyper = HyperData(
(np.transpose(x_test, (0, 3, 1, 2)).astype("float32"), y_test), None)
kwargs = {'num_workers': 1, 'pin_memory': True}
train_loader = torch.utils.data.DataLoader(train_hyper,
batch_size=args.tr_bsize,
shuffle=True,
**kwargs)
test_loader = torch.utils.data.DataLoader(test_hyper,
batch_size=args.te_bsize,
shuffle=False,
**kwargs)
return train_loader, test_loader, numberofclass, bands
def load_hyper(args):
if args.dataset not in ["UH", "DIP", "DUP"]:
acrstate = None if args.random_state == None else args.random_state+args.idtest
pixelsO, labelsO, numclass = auxil.loadData(args.dataset, num_components=args.components, rand_state=acrstate)
pixelsO, labelsO = auxil.createImageCubes(pixelsO, labelsO, windowSize=args.spatialsize, removeZeroLabels = False)
pixels = pixelsO[labelsO!=0]
labels = labelsO[labelsO!=0] - 1
bands = pixels.shape[-1]; numberofclass = len(np.unique(labels))
pixels = pixels.reshape((pixels.shape[0], args.spatialsize, args.spatialsize, bands, 1))
pixelsO = pixelsO.reshape((pixelsO.shape[0], args.spatialsize, args.spatialsize, bands, 1))
x_train, x_test, y_train, y_test = auxil.split_data(pixels, labels, args.tr_percent, rand_state=acrstate)
del pixels, labels
else:
x_train, x_test, y_train, y_test, pixelsO, labelsO, bands, numberofclass = auxil.loadDataFIX(args)
return (x_train, y_train), (x_test, y_test), (pixelsO, labelsO), numberofclass, bands
def load_hyper(args):
data, labels, numclass = auxil.loadData(args.dataset, num_components=args.components)
pixels, labels = auxil.createImageCubes(data, labels, windowSize=args.spatialsize, removeZeroLabels = True)
bands = pixels.shape[-1]; numberofclass = len(np.unique(labels))
x_train, x_test, y_train, y_test = auxil.split_data(pixels, labels, args.tr_percent)
if args.use_val: x_val, x_test, y_val, y_test = auxil.split_data(x_test, y_test, args.val_percent)
del pixels, labels
if args.p != 0: transform_train = transform.RandomErasing(probability = args.p, sh = args.sh, r1 = args.r1,)
else: transform_train = None
train_hyper = HyperData((np.transpose(x_train, (0, 3, 1, 2)).astype("float32"),y_train), transform_train)
test_hyper = HyperData((np.transpose(x_test, (0, 3, 1, 2)).astype("float32"),y_test), None)
if args.use_val: val_hyper = HyperData((np.transpose(x_val, (0, 3, 1, 2)).astype("float32"),y_val), None)
kwargs = {'num_workers': 1, 'pin_memory': True}
train_loader = torch.utils.data.DataLoader(train_hyper, batch_size=args.tr_bsize, shuffle=True, **kwargs)
test_loader = torch.utils.data.DataLoader(test_hyper, batch_size=args.te_bsize, shuffle=False, **kwargs)
if args.use_val: val_loader = torch.utils.data.DataLoader(val_hyper, batch_size=args.val_bsize, shuffle=False, **kwargs)
else: val_loader = None
return train_loader, val_loader, test_loader, numberofclass, bands