GRADE / environment.txt
Bin-0815's picture
Release all GRADE models, checkpoints, and reviewed evaluation code
0e150d6 verified
Raw History Blame Contribute Delete
1.33 kB
# Minimal inference, metric, and quantitative-reproduction environment.
# Python 3.11 is required. Install the CUDA-matched PyTorch wheel
# for the host before (or together with) this file; see README.md.
# Core CUDA / model runtime (stable counterparts of the server environment).
torch==2.7.0
torchvision==0.22.0
torchmetrics==1.9.0
accelerate==1.11.0
diffusers==0.39.0
transformers==5.14.1
timm==1.0.28
safetensors==0.8.0
huggingface_hub==1.24.0
# Data loading and model utilities.
numpy==1.26.4
scipy==1.12.0
opencv-python>=4.10,<5
Pillow==11.1.0
PyYAML==6.0.2
einops==0.8.1
noise==1.2.2
tqdm==4.67.1
# Metrics and quantitative paper reproduction.
# The LPIPS evaluator downloads TorchVision's AlexNet weights on first use.
# Warm the TorchVision cache online before running metrics offline; see README.md.
pandas==2.2.3
matplotlib==3.10.0
open3d==0.19.0
dash==3.2.0 # Open3D imports its visualization module during 3D metric startup.
plotly==6.3.1 # Imported by Open3D's visualization package during 3D metrics.
# Pinned Depth Anything 3 source used by the DA3 baseline.
addict==2.4.0 # Imported by the pinned DA3 source but absent from its metadata.
moviepy==1.0.3 # DA3 imports moviepy.editor; MoviePy 2 removed it.
git+https://github.com/ByteDance-Seed/Depth-Anything-3.git@3d835ec1a5802d64a8b8b15f817a1ab54809bfe4