# ResDiffFRG

This repository provides a working demo of **ResDiffFRG: Residual Diffusion for Multiple Appropriate Facial Reaction Generation**

[Paper](https://doi.org/10.48550/arXiv.2609.33749)

## Requirements

- Python >= 3.10
- PyTorch
- NumPy

A CUDA GPU is recommended.

## Running the demo

From this directory, run:

    python sample_run.py

The script loads the anonymised speaker example provided in the sample_data folder. It generates 10 25-dimensinoal facial reactions using ResDiffFRG and saves the output as a .npy file to the results folder.