stylegan_op_upfirdn2d.py

from collections import abc

import torch
from torch.nn import functional as F


def upfirdn2d(inputs, kernel, up=1, down=1, pad=(0, 0)):
    if not isinstance(up, abc.Iterable):
        up = (up, up)

    if not isinstance(down, abc.Iterable):
        down = (down, down)

    if len(pad) == 2:
        pad = (pad[0], pad[1], pad[0], pad[1])

    return upfirdn2d_native(inputs, kernel, *up, *down, *pad)


def upfirdn2d_native(
    inputs, kernel, up_x, up_y, down_x, down_y, pad_x0, pad_x1, pad_y0, pad_y1
):
    _, channel, in_h, in_w = inputs.shape
    inputs = inputs.reshape(-1, in_h, in_w, 1)

    _, in_h, in_w, minor = inputs.shape
    kernel_h, kernel_w = kernel.shape

    out = inputs.view(-1, in_h, 1, in_w, 1, minor)
    out = F.pad(out, [0, 0, 0, up_x - 1, 0, 0, 0, up_y - 1])
    out = out.view(-1, in_h * up_y, in_w * up_x, minor)

    out = F.pad(
        out, [0, 0, max(pad_x0, 0), max(pad_x1, 0), max(pad_y0, 0), max(pad_y1, 0)]
    )
    out = out[
          :,
          max(-pad_y0, 0): out.shape[1] - max(-pad_y1, 0),
          max(-pad_x0, 0): out.shape[2] - max(-pad_x1, 0),
          :,
          ]

    out = out.permute(0, 3, 1, 2)
    out = out.reshape(
        [-1, 1, in_h * up_y + pad_y0 + pad_y1, in_w * up_x + pad_x0 + pad_x1]
    )
    w = torch.flip(kernel, [0, 1]).view(1, 1, kernel_h, kernel_w)
    out = F.conv2d(out, w)
    out = out.reshape(
        -1,
        minor,
        in_h * up_y + pad_y0 + pad_y1 - kernel_h + 1,
        in_w * up_x + pad_x0 + pad_x1 - kernel_w + 1,
    )
    out = out.permute(0, 2, 3, 1)
    out = out[:, ::down_y, ::down_x, :]

    out_h = (in_h * up_y + pad_y0 + pad_y1 - kernel_h + down_y) // down_y
    out_w = (in_w * up_x + pad_x0 + pad_x1 - kernel_w + down_x) // down_x

    return out.view(-1, channel, out_h, out_w)