requirements.txt

# Core dependencies
Pillow>=10.0.0
DateTime>=5.0
matplotlib>=3.7.0
plyfile>=1.1.0
HTML4Vision>=0.5.0
timm>=1.0.0
tensorboardX>=2.6.0
imgaug>=0.4.0
iopath>=0.1.0
imagecorruptions>=1.1.0
numpy>=1.24.0
lpips>=0.1.4
gradio>=4.0.0
opencv-python>=4.9.0
scipy>=1.14.0
scikit-image>=0.24.0
pytorch-lightning>=2.5.0
einops>=0.8.0
omegaconf>=2.3.0
ftfy>=6.3.0

# PyTorch ecosystem (install via setup_env.sh only)
# torch, torchvision, torchaudio installed separately with the correct CUDA wheels

# 3D and computer vision
open3d>=0.19.0
# Pin kornia to avoid flash-attn dependency path
kornia>=0.7.2,<0.8.0
# pytorch3d>=0.7.8  # Commented out due to version conflicts - will use fallback renderer
trimesh>=4.0.0

# AI/ML frameworks
diffusers>=0.34.0
transformers>=4.30.0
accelerate>=0.30.0
huggingface-hub>=0.30.0

# Memory optimization (optional): install xformers via setup_env.sh with CUDA-specific wheels

# Computer vision and ML
mmcv>=2.2.0
mmengine>=0.10.0
einops>=0.8.0

# Optional: Install VGGT directly from GitHub
# uv pip install "git+https://github.com/facebookresearch/vggt.git"
# uv pip install "git+https://github.com/isl-org/Depth-Anything.git"

# Rerun for visualization
rerun-sdk>=0.24.0

# VGGT dependencies (installed separately)
# VGGT requires manual setup: git clone https://github.com/facebookresearch/vggt.git ../vggt
# Then create missing __init__.py files as shown in setup_rtx5080_complete.sh