def __getitem__(self, idx):
#if self.use_npy:
# input = self.dsets['low'][idx]
# target = self.dsets['high'][idx]
input = load_img(self.dsets['low'][idx])
target = load_img(self.dsets['high'][idx])
if len(input.shape) == 2:
input = input.reshape(1, input.shape[0], input.shape[1])
target = target.reshape(1, target.shape[0], target.shape[1])
else:
input = np.transpose(input, (2, 0, 1))
target = np.transpose(target, (2, 0, 1))
input = torch.from_numpy(input).type(torch.FloatTensor)
target = torch.from_numpy(target).type(torch.FloatTensor)
return input, target
def prepare_data(args):
print("Preparing DIV2K dataset ...")
h5file_path = "{}/DIV2K_np_test_{}.h5".format(args.h5file_dir, args.test_sigma)
h5 = h5py.File(h5file_path, 'w')
h_group = h5.create_group('h')
l_group = h5.create_group('l')
h_list = sorted(glob.glob(args.test_h + "*.png"))
with tqdm(total=len(h_list)) as t:
t.set_description("H & L")
for i, path in enumerate(h_list):
img = img2np(load_img(path))
h_group.create_dataset(str(i), data=img)
l_group.create_dataset(str(i), data=add_noise(img, float(args.test_sigma), train=False))
t.update()
h5.close()
print("Prepare successfully")