Method forward

data/shift_augmentation.py:20–25

(self, x)

                                         CenterCrop(size=n_size + (self.shift_max if np.random.random() < p else 0))(x)))

    def forward(self, x):
        # X is shaped: (C, H, W)
        aug_x = torch.zeros((x.shape[0], self.n_size, self.n_size))
        for i in range(x.shape[0]):
            aug_x[i, :, :] = self.chanel_shifter(x[[i], :, :])[0,:,:]
        return aug_x