pyproject.toml

[build-system]
requires = ["setuptools>=68", "wheel"]
build-backend = "setuptools.build_meta"

[project]
name = "egogeneval"
version = "0.1.0"
description = "One-command evaluation tooling for the EgoGenEval benchmark"
requires-python = ">=3.10,<3.13"
license = {text = "MIT"}
authors = [
  {name = "Yilin Long"},
  {name = "Chenming Zhu"},
  {name = "Zitang Gou"},
  {name = "Jingli Lin"},
  {name = "Tai Wang"},
]
dependencies = [
  "jsonschema>=4.22,<5",
  "Pillow>=10,<12",
  "PyYAML>=6,<7",
]

[project.optional-dependencies]
prepare = [
  "numpy>=1.26,<2.1",
  "opencv-python-headless>=4.9,<5",
  "imageio>=2.34,<3",
  "h5py>=3.10,<4",
  "tqdm>=4.66,<5",
]
scannetpp = [
  "numpy==2.0.2",
  "opencv-python-headless==4.10.0.84",
  "pyrender==0.1.45",
  "PyOpenGL==3.1.0",
  "trimesh==5.1.0",
  "scipy>=1.11,<2",
  "Pillow==11.3.0",
]
data = [
  "datasets>=2.19,<4",
]
# Decoding a video model's per-step clips. Not needed when steps are already
# extracted stills or frame directories.
video = [
  "imageio>=2.34,<3",
  "imageio-ffmpeg>=0.5,<1",
]
full = [
  "numpy>=1.26,<2.1",
  "scipy>=1.11,<2",
  "Pillow>=10,<12",
  "opencv-python-headless>=4.9,<5",
  "imageio>=2.34,<3",
  "imageio-ffmpeg>=0.5,<1",
  "h5py>=3.10,<4",
  "tqdm>=4.66,<5",
  "torch>=2.4,<2.8",
  "torchvision>=0.19,<0.23",
  "transformers>=4.51,<5",
  "accelerate>=1.0,<2",
  "huggingface-hub>=0.27,<2",
  "safetensors>=0.4,<1",
  "openai>=1.50,<2",
  "qwen-vl-utils>=0.0.8",
  "scikit-image>=0.24,<1",
]
[project.scripts]
egogeneval = "egogeneval.cli:entrypoint"

[tool.setuptools.packages.find]
where = ["src"]

[tool.setuptools.package-data]
egogeneval = [
  "resources/*.json",
  "resources/schemas/*.json",
  "official/VENDORED_SHA256SUMS",
]