Installation

Bitbox itself has minimum requirements, but it relies on face/body backends to generate expression and movement signals. These backends usually have more requirements. We highly recommend using our Docker images to install these backends, as installing them from source code may prove difficult for some. For currently supported backends, you need NVIDIA GPUs. New backends with CPU support are coming soon.

Installing Backends

The current version of Bitbox supports two face backends, namely 3DI and 3DI-lite. While 3DI provides more detailed outputs (e.g., a full 3D model of the face), 3DI-lite is much faster and more robust to occlusions. If your images don't have significant occlusions and you don't need a faster solution, we recommend using 3DI.

We recommend using Docker for installing all backends at once with no extra steps.

Using Docker is usually straightforward. However, 3DI requires downloading an external face model, which involves individually registering and requesting access, and updating our image with this model.

  1. Download the Dockerfile. We have two options for CUDA 11.8.0 or CUDA 12.2.2. Select the one that is most compatible with your NVIDIA GPU.

  2. Register and download the face model. This is an automated process, and you should immediately get the download link in your email. Extract the model you downloaded and locate the file 01_MorphableModel.mat .

  3. Place the Dockerfile and the face model (01_MorphableModel.mat) in the same directory

  4. Within this directory, run the following command to build the image

    docker build -t bitbox:latest .

    The parameter bitbox:latest is the name of the image to be created. You can change it if you wish. Please don't forget the . at the end.

Installing Python Library

Once you installed the backends, you can install the stable version of the Bitbox Python library using pip or the latest version directly from GitHub.

pip3 install bitbox