Skip to content

Latest commit

 

History

History
60 lines (38 loc) · 1.78 KB

README.md

File metadata and controls

60 lines (38 loc) · 1.78 KB
Raw

Channel-aware GAN Inversion for Semantic Communication

This repository contains the implementation of a channel-aware GAN inversion method designed to extract meaningful semantic information from original inputs and map it into a channel-correlated latent space, which can eliminates the necessity for additional channel encoder and decoder.

Prerequisites

Make sure you have the following dependencies installed:

  • Python (>=3.6)
  • PyTorch
  • NumPy
  • Pillow
  • imageio
  • tqdm
  • lpips
  • piq
  • pytorch-msssim

You can install the required packages using the following command:

pip install -r requirements.txt

Getting Started

  1. Clone the repository:
git clone https://github.com/recusant7/GAN_SeCom.git
cd GAN_SeCom
  1. Download the pre-trained model checkpoint:
wget https://github.com/seasonSH/SemanticStyleGAN/releases/download/1.0.0/CelebAMask-HQ-512x512.pt -O pretrained/CelebAMask-HQ-512x512.pt

Running the Code

Run the inversion code with the following command:

python main.py --ckpt pretrained/CelebAMask-HQ-512x512.pt --outdir results/inversion --dataset ./data/examples --size 512 --batch_size 8 --snr_db 15

Make sure to adjust the arguments based on your requirements. You can find a description of the available arguments in the script.

Results

Reconstructed Images The results, including reconstructed images and log files, will be saved in the specified output directory (--outdir). Check the log files for average PSNR, MS-SSIM, and LPIPS.

Acknowledgments

  • This code is based on SemanticGAN and SemanticStyleGAN. We extend our sincere thanks to the authors of these projects for their valuable works.