Skip to content

INSTALL.md

Latest commit

 

History

History
114 lines (79 loc) · 2.75 KB

INSTALL.md

File metadata and controls

114 lines (79 loc) · 2.75 KB

Installation

Tested on Rocky Linux 8.10, NVIDIA Ampere A40, CUDA 12.1, Python 3.10.

1. Clone Repository

git clone https://github.com/UVA-Computer-Vision-Lab/LabelAny3D.git
cd LabelAny3D
export EXT_DIR=$(pwd)/external

2. Create Environment

conda create -n la3d python=3.10
conda activate la3d

3. Install PyTorch and Dependencies

pip install torch==2.2.2 torchvision==0.17.2 --index-url https://download.pytorch.org/whl/cu121
pip install -r requirements.txt

4. Install Additional Packages

pip install git+https://github.com/facebookresearch/pytorch3d.git@055ab3a --no-build-isolation
pip install git+https://github.com/yaojin17/detectron2.git --no-build-isolation
pip install pycocotools==2.0 --no-build-isolation

5. Install External Dependencies

MoGe (Depth Estimation)

cd $EXT_DIR/MoGe && pip install -r requirements.txt

DepthPro (Metric Depth)

cd $EXT_DIR/ml-depth-pro && pip install -e .

TRELLIS (3D Reconstruction)

Requires GCC 11+.

export CC=$(which gcc)
export CXX=$(which g++)
cd $EXT_DIR/TRELLIS && . ./setup.sh --basic --xformers --diffoctreerast --spconv --mipgaussian --nvdiffrast

flash-attn

Download pre-built wheel for your Python/PyTorch/CUDA version from flash-attention releases.

Example for Python 3.10, PyTorch 2.2, CUDA 12:

wget https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.2cxx11abiFALSE-cp310-cp310-linux_x86_64.whl
pip install flash_attn-2.7.4.post1+cu12torch2.2cxx11abiFALSE-cp310-cp310-linux_x86_64.whl
rm flash_attn-2.7.4.post1+cu12torch2.2cxx11abiFALSE-cp310-cp310-linux_x86_64.whl

kaolin

Build from source (required for older glibc systems):

git clone --recursive https://github.com/NVIDIAGameWorks/kaolin.git /tmp/kaolin
cd /tmp/kaolin && git checkout v0.17.0
pip install . --no-build-isolation

InvSR (Image Enhancement)

cd $EXT_DIR/InvSR && pip install -e . --no-deps

6. Download Checkpoints

cd $EXT_DIR/checkpoints
./download.sh

This downloads:

  • DepthPro - metric depth estimation (~1.8GB)
  • InvSR - image super-resolution (~130MB)
  • Amodal Completion - complete occluded regions (~3.3GB)

Other models (TRELLIS, MoGe) are auto-downloaded from HuggingFace at runtime.

7. Blender (Optional, for Visualization)

Install Blender 3.6+ and add trimesh:

blender --background --python-expr "import subprocess, sys; subprocess.check_call([sys.executable, '-m', 'pip', 'install', 'trimesh'])"

Verify Installation

cd src
python -c "import torch; print(f'PyTorch: {torch.__version__}, CUDA: {torch.cuda.is_available()}')"