Salient Co-Speech Gesture Generation

A repository for generating salient co-speech gesture. This is the repository for paper Salient Co-Speech Gesture Synthesizing with Discrete Motion Representation.

framework

Quick Start

Installation

conda create -f environment.yml python=3.8

Run

  1. Download dataset to data/baijia_all
  2. Run train_vae.py -c data/motion_vqvae.yml to train Motion Auto-Encoder.
  3. Run train_transformer.py -c data/motion_vqvae.yml to train Speech-Gesture Mapping network.
  4. Run evaluate_transformer.py to evaluate FGD and MVD.
  5. Run inference_transformer.py to evaluate SMS.

Data

We leverage lmdb to store the dataset. The data can be downloaed at link, and then run cat xa* > data.mdb.

Project Overview

TODO