python package for self-attention gan implemented as extension of PyTorch nn.Module. paper -> https://arxiv.org/abs/1805.08318

Also includes generic layers for image based attention mechanism. Includes a Full-Attention layer as proposed by in another project of mine here

This is a python package availbale at the pypi.org. So, installation is fairly straightforward. This package depends on a suitable GPU version of torch and torch-vision for your architecture. So, please download suitable pytorch prior to installing this package. Follow the instructions at pytorch.org to install your version of PyTorch.

Install with following commands:

$ workon [your virtual environment] 
$ pip install attn-gan-pytorch

some celeba samples generated using this code for the fagan architecture:

Also, this repo contains the code for using this package to build the SAGAN architecture as mentioned in the paper. Please refer the samples/ directory for this.

Please feel free to open PRs here if you train on other datasets using this package. Suggestions / Issues / Contributions are most welcome.

Best regards,
@akanimax :)