Skip to content

INSTALL.md

Latest commit

 

History

History
93 lines (63 loc) · 2.7 KB

INSTALL.md

File metadata and controls

93 lines (63 loc) · 2.7 KB

Installation Guide

Prerequisites

  • Python 3.11 (recommended via conda or pyenv)
  • CUDA 12.x compatible GPU with 16+ GB VRAM (tested on RTX 4090)
  • ffmpeg — required for video compilation (winget install ffmpeg on Windows, brew install ffmpeg on macOS)
  • Git — for cloning dependencies

1. Create and Activate Environment

conda create -n sam_3d_body python=3.11 -y
conda activate sam_3d_body

2. Install PyTorch

Install PyTorch with CUDA support following the official instructions.

# Example for CUDA 12.1:
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

3. Install Python Dependencies

pip install pytorch-lightning pyrender opencv-python yacs scikit-image einops timm dill pandas rich \
    hydra-core hydra-submitit-launcher hydra-colorlog pyrootutils webdataset networkx==3.2.1 roma \
    joblib seaborn wandb appdirs ffmpeg cython jsonlines pytest loguru optree fvcore black \
    pycocotools tensorboard huggingface_hub tqdm imageio imageio-ffmpeg yt-dlp

4. Install Detectron2

pip install 'git+https://github.com/facebookresearch/detectron2.git@a1ce2f9' --no-build-isolation --no-deps

On Windows, you may need Visual Studio Build Tools. See build_detectron2.bat for a helper script.

5. Install MoGe (FOV Estimation)

pip install git+https://github.com/microsoft/MoGe.git

6. Install SAM3 (Optional — for SAM3 detector)

git clone https://github.com/facebookresearch/sam3.git
cd sam3
pip install -e .
pip install decord psutil

Getting Model Checkpoints

We host model checkpoints on Hugging Face. Available models:

Note: You must request access on the HuggingFace repos above before downloading.

# Authenticate
huggingface-cli login

# Download all models (SAM 3D Body + ViTDet + MoGe2 + SAM2)
python download_models.py

# Or download only the primary model:
python download_models.py --skip-sam2 --skip-vitdet --skip-moge

API Keys

Copy the template and fill in your keys:

cp .env.example .env
  • GROQ_API_KEY — Required for voice control in the Avatar Studio (get one here)
  • HF_TOKEN — Required to download gated model checkpoints (get one here)