| # Optional overrides. Copy to env.local.sh and edit. | |
| # | |
| # Self-contained layout (default, no edit needed): | |
| # benchmarks/datasets_portable/ benchmarks/assets/ | |
| # benchmarks/manifests/ videos/<bench>/<method>/ | |
| # results/ deps/ -> evaluations repo | |
| # | |
| # One-time: bash scripts/setup_deps.sh /path/to/projects/evaluations | |
| # Metric implementations (VBench, VideoAlign, obj_dino, internvl) | |
| # export EVAL_DEPS_ROOT=/path/to/evaluations | |
| # Conda envs or absolute python (optional) | |
| # export PYTHON_OBJ_DINO=... | |
| # export PYTHON_VBENCH=... | |
| # export PYTHON_TVA=... | |
| # export PYTHON_INTERNVL=... | |
| # Model weights for TVA / InternVL | |
| export TVA_MODEL_PATH="${TVA_MODEL_PATH:-/path/to/VideoReward}" | |
| export INTERNVL_MODEL_PATH="${INTERNVL_MODEL_PATH:-/path/to/InternVL3_5-38B}" | |
| export VBENCH_CACHE_DIR="${VBENCH_CACHE_DIR:-${EVAL_DEPS_ROOT:-./deps}/VBench/weghts}" | |
| # Override default bundled videos (else uses videos/<benchmark>/<method>/) | |
| # export VIDEO_PATH=... | |
| # export OUTPUT_DIR=... | |