Vertices to Flame

This repo contains a 3d vertices to flame parameters (pose and expression) model. This model convert vertices to blend shapes, flame expression blend shapes and jaw rotation parameters. It works as a flame inverter to extract back the parameters from the vertices

Environment

• Linux
• Python 3.6+
• Pytorch 2.0.1

Other necessary packages:

pip install -r requirements.txt

• unzip (for dataset extraction)

Usage

import torch
from vertices2flame import FlameInverter
inverter = FlameInverter(from_pretrained=True)
vert = torch.zeros((1, 1, 15069))
pose, exp = inverter(vert)

Dataset Preparation

Download "FLAME_sample.ply" from voca and put it in dataset/model.

VoxCeleb

Request the Flame model from https://flame.is.tue.mpg.de/. Place the download files flame_static_embedding.pkl, flame_dynamic_embedding.npy and generic_model.pkl in dataset/model Download the dataset from the bucket:

 python download_dataset.py audio2face /path/to/credential.json

Training

The training operation shares a similar command:

sh train.sh FLAME_1 config/train.yaml vox 

Acknowledgement

We heavily borrow the code from Codetalker,