Alias-Free-Torch

Simple torch module implementation of Alias-Free GAN.

This repository including

• Alias-Free GAN style lowpass sinc filter @filter.py

• Alias-Free GAN style up/downsample @resample.py

• Alias-Free activation @act.py

• and test codes @./test

Note: Since this repository is unofficial, filter and upsample could be different with official implementation.

Still working! If you notice some error or typo, please open new issue! v0.0.8 is TESTED

UPDATE: You can download alias-free-torch from pip

python -m pip install alias-free-torch

Requirements

Due to torch.kaiser_window and torch.i0 are implemeted after 1.7.0, our repository need torch>=1.7.0.

• Pytorch>=1.7.0

For custom torch users, pip will not check torch version.

TODO

• 2d sinc filter
• 2d resample
• devide 1d and 2d modules
• pip packaging
• rewrite upsample
• Upsample pad size issue
• (Upsample) support calculation for [B,C,T/(H,W)] (now only supports [B,T/(H,W)] or [B,1,T/(H,W)])
• set filter as register buffer
• (Downsample & Filter) support calculation for [B,C,T/(H,W)] (now only supports [B,T/(H,W)] or [B,1,T/(H,W)])
• provide loadable ckpt for lower version of torch
• documentation

Test results 1d

Filter sine	Filter noise

upsample	downsample

Test results 2d

Filter L1 norm sine	Filter noise

upsample	downsample

References

• Alias-Free GAN
• adefossez/julius
• A. V. Oppenheim and R. W. Schafer. Discrete-Time Signal Processing. Pearson, International Edition, 3rd edition, 2010

Acknowledgement

This work is done at MINDsLab Inc.

Thanks to teammates at MINDsLab Inc.

• Hyoungkyu Song
• Kang-wook Kim
• Dongho Choi
• Sang Hoon Woo