Search

nn.Pixel Shuffle

class torch.nn.PixelShuffle(upscale_factor)
(∗,  C×r2,  H,  W)(*,\; C\times r^2,\; H, \; W) → (∗,  C,  H×r,  W×r)(*,\; C,\; H\times r, \; W\times r) 로 요소들을 rearrange해준다
r은 upscale factor이다
stride 1/r로 efficient sub-pixel conv를 적용할 때 유용하다
PixelShuffle - PyTorch 1.11.0 documentation
Join the PyTorch developer community to contribute, learn, and get your questions answered.
https://pytorch.org/docs/stable/generated/torch.nn.PixelShuffle.html
Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network
Recently, several models based on deep neural networks have achieved great success in terms of both reconstruction accuracy and computational performance for single image super-resolution. In these methods, the low resolution (LR) input image is upscaled to the high resolution (HR) space using a single filter, commonly bicubic interpolation, before reconstruction.
https://arxiv.org/abs/1609.05158
이 논문에서 나온 방법이다
원본이 (H, W, 3)이미지라면 이를 마지막 레이어에서 (H, W, 3 x r2r^2) 로 만들어서 순서대로 조합해 (Hxr, Wxr, 3) 의 이미지를 만들어내는 것이다
Made with