evaluation_code/scripts/eval_mme_cof.sh

#!/usr/bin/env bash
set -euo pipefail
REPO_ROOT=$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)
export PYTHONPATH="$REPO_ROOT/src${PYTHONPATH:+:$PYTHONPATH}"

if [[ $# -lt 3 || $# -gt 4 ]]; then
  echo "Usage: $0 MME_METADATA_JSON VIDEOS_DIR OUTPUT_JSONL [SEEDS]" >&2
  exit 2
fi

METADATA_JSON=$1
VIDEOS_DIR=$2
OUTPUT_JSONL=$3
SEEDS=${4:-0,1,2}

python -m vwg_bench.cli evaluate-mme-cof \
  --metadata-json "$METADATA_JSON" \
  --videos-dir "$VIDEOS_DIR" \
  --output "$OUTPUT_JSONL" \
  --seeds "$SEEDS"

python -m vwg_bench.cli summarize-mme-cof \
  --results "$OUTPUT_JSONL" \
  --output "${OUTPUT_JSONL%.jsonl}.summary.json"