import torch.nn.functional as F


class InputPadder:
    """Pads images such that dimensions are divisible by divisor"""

    def __init__(self, dims, divisor=16):
        self.ht, self.wd = dims[-2:]
        pad_ht = (((self.ht // divisor) + 1) * divisor - self.ht) % divisor
        pad_wd = (((self.wd // divisor) + 1) * divisor - self.wd) % divisor
        self._pad = [
            pad_wd // 2,
            pad_wd - pad_wd // 2,
            pad_ht // 2,
            pad_ht - pad_ht // 2,
        ]

    def pad(self, *inputs):
        return [F.pad(x, self._pad, mode="replicate") for x in inputs]

    def unpad(self, x):
        ht, wd = x.shape[-2:]
        c = [self._pad[2], ht - self._pad[3], self._pad[0], wd - self._pad[1]]
        return x[..., c[0] : c[1], c[2] : c[3]]