Datasets:

ProAudience
/

agenthoi-eval

Modalities:

agenthoi-eval / scripts /run_all_metrics.sh

Upload folder using huggingface_hub

271f3e5 verified 6 months ago

6.2 kB

#!/bin/bash
# Run all HOI evaluation metrics (self-contained paths under evaluation/).
#
# Prefer: bash scripts/run_eval.sh <benchmark> <method>
# Or set VIDEO_PATH + source eval_common.sh, then run this script.
#
# Requires: bash scripts/setup_deps.sh /path/to/evaluations  (once)
# Optional: env.local.sh for PYTHON_* / model weights

set -e

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
# shellcheck source=eval_common.sh
source "${SCRIPT_DIR}/eval_common.sh"

if [ -z "${EVAL_DEPS_ROOT:-}" ] || [ ! -d "${EVAL_DEPS_ROOT}/VBench" ]; then
  echo "[run_all] ERROR: metric code not found." >&2
  echo "  Run: bash scripts/setup_deps.sh /path/to/projects/evaluations" >&2
  echo "  Or: export EVAL_DEPS_ROOT=/path/to/evaluations" >&2
  exit 1
fi

VIDEO_PATH="${VIDEO_PATH:?Set VIDEO_PATH or use: bash scripts/run_eval.sh <bench> <method>}"
OUTPUT_DIR="${OUTPUT_DIR:-${RESULTS_DIR}/$(basename "${VIDEO_PATH}")_eval}"

OBJ_DINO_JSON="${OBJ_DINO_JSON:-${MANIFESTS_DIR}/obj_dino_crosstest.json}"
TVA_JSON="${TVA_JSON:-${MANIFESTS_DIR}/tva_crosstest.json}"
INTERNVL_JSON="${INTERNVL_JSON:-${DATASETS_PORTABLE}/hoi_crosstest_20251107.json}"

SKIP_OBJ_DINO="${SKIP_OBJ_DINO:-0}"
SKIP_VBENCH="${SKIP_VBENCH:-0}"
SKIP_TVA="${SKIP_TVA:-0}"
SKIP_INTERNVL="${SKIP_INTERNVL:-0}"

PYTHON_OBJ_DINO="${PYTHON_OBJ_DINO:-}"
CONDA_OBJ_DINO="${CONDA_OBJ_DINO:-dino_obj}"
PYTHON_VBENCH="${PYTHON_VBENCH:-}"
CONDA_VBENCH="${CONDA_VBENCH:-dl-main}"
PYTHON_TVA="${PYTHON_TVA:-}"
CONDA_VIDEOREWARD="${CONDA_VIDEOREWARD:-VideoReward}"
PYTHON_INTERNVL="${PYTHON_INTERNVL:-}"
CONDA_INTERNVL="${CONDA_INTERNVL:-internvl}"

TVA_MODEL_PATH="${TVA_MODEL_PATH:-/path/to/VideoReward}"
INTERNVL_MODEL_PATH="${INTERNVL_MODEL_PATH:-/path/to/InternVL3_5-38B}"
TVA_NUM_GPUS="${TVA_NUM_GPUS:-8}"
export VBENCH_CACHE_DIR="${VBENCH_CACHE_DIR:-${EVAL_DEPS_ROOT}/VBench/weghts}"

mkdir -p "${OUTPUT_DIR}"

# InternVL needs absolute ref-image paths when using datasets_portable/
INTERNVL_JSON_RESOLVED="${OUTPUT_DIR}/.internvl_dataset.json"
python3 "${SCRIPT_DIR}/absolutize_dataset_json.py" \
  --eval-root "${EVAL_ROOT}" \
  --input "${INTERNVL_JSON}" \
  --output "${INTERNVL_JSON_RESOLVED}"
INTERNVL_JSON="${INTERNVL_JSON_RESOLVED}"

_run_python() {
  local py="$1" conda_env="$2"
  shift 2
  if [ -n "${py}" ]; then
    "${py}" "$@"
  else
    conda run -n "${conda_env}" --no-capture-output --live-stream python "$@"
  fi
}

echo "[run_all] EVAL_ROOT=${EVAL_ROOT}"
echo "[run_all] EVAL_DEPS_ROOT=${EVAL_DEPS_ROOT}"
echo "[run_all] VIDEO_PATH=${VIDEO_PATH}"
echo "[run_all] OUTPUT_DIR=${OUTPUT_DIR}"

if [ "${SKIP_OBJ_DINO}" = "0" ]; then
  echo "[run_all] ----- 1/4 Object-DINO -----"
  OBJ_DINO_SAVE="${OUTPUT_DIR}/obj_dino_result.json"
  _run_python "${PYTHON_OBJ_DINO}" "${CONDA_OBJ_DINO}" \
    "${EVAL_DEPS_ROOT}/obj_dino/obj_dino.py" \
    --video_path "${VIDEO_PATH}" \
    --json_list "${OBJ_DINO_JSON}" \
    --save_path "${OBJ_DINO_SAVE}"
  echo "[run_all] Object-DINO -> ${OBJ_DINO_SAVE}"
else
  echo "[run_all] SKIP Object-DINO"
fi

if [ "${SKIP_VBENCH}" = "0" ]; then
  echo "[run_all] ----- 2/4 VBench -----"
  VBENCH_OUT="${OUTPUT_DIR}/vbench_results"
  if [ -n "${PYTHON_VBENCH}" ]; then
    (cd "${EVAL_DEPS_ROOT}/VBench" && "${PYTHON_VBENCH}" evaluate.py \
      --videos_path "${VIDEO_PATH}" \
      --output_path "${VBENCH_OUT}" \
      --mode custom_input \
      --dimension motion_smoothness dynamic_degree aesthetic_quality overall_consistency)
  else
    conda run -n "${CONDA_VBENCH}" --no-capture-output --live-stream \
      bash -c "cd ${EVAL_DEPS_ROOT}/VBench && python evaluate.py \
        --videos_path \"${VIDEO_PATH}\" \
        --output_path \"${VBENCH_OUT}\" \
        --mode custom_input \
        --dimension motion_smoothness dynamic_degree aesthetic_quality overall_consistency"
  fi
  echo "[run_all] VBench -> ${VBENCH_OUT}"
else
  echo "[run_all] SKIP VBench"
fi

if [ "${SKIP_TVA}" = "0" ]; then
  echo "[run_all] ----- 3/4 TVA (VideoAlign) -----"
  TVA_SAVE="${OUTPUT_DIR}/TVA_result.json"
  if [ -f "${EVAL_DEPS_ROOT}/VideoAlign/TVA_metrics_hzy_v2.py" ]; then
    if [ -n "${PYTHON_TVA}" ]; then
      (cd "${EVAL_DEPS_ROOT}/VideoAlign" && "${PYTHON_TVA}" TVA_metrics_hzy_v2.py \
        --video_path "${VIDEO_PATH}" \
        --json_list "${TVA_JSON}" \
        --save_path "${TVA_SAVE}" \
        --gpu_id 0)
    else
      conda run -n "${CONDA_VIDEOREWARD}" --no-capture-output --live-stream \
        bash -c "cd ${EVAL_DEPS_ROOT}/VideoAlign && python TVA_metrics_hzy_v2.py \
          --video_path \"${VIDEO_PATH}\" \
          --json_list \"${TVA_JSON}\" \
          --save_path \"${TVA_SAVE}\" \
          --gpu_id 0"
    fi
  else
    _run_python "${PYTHON_TVA}" "${CONDA_VIDEOREWARD}" \
      "${EVAL_DEPS_ROOT}/VideoAlign/TVA_metrics.py" \
      --model_path "${TVA_MODEL_PATH}" \
      --video_path "${VIDEO_PATH}" \
      --json_list "${TVA_JSON}" \
      --save_path "${TVA_SAVE}" \
      --num_gpus "${TVA_NUM_GPUS}"
  fi
  echo "[run_all] TVA -> ${TVA_SAVE}"
else
  echo "[run_all] SKIP TVA"
fi

if [ "${SKIP_INTERNVL}" = "0" ]; then
  echo "[run_all] ----- 4/4 InternVL -----"
  INTERNVL_OUT="${OUTPUT_DIR}/internvl"
  INTERNVL_NUM_PROCS=8
  for _gpu in 0 1 2 3 4 5 6 7; do
    _cur="${_gpu}"
    (
      export CUDA_VISIBLE_DEVICES="${_gpu}"
      _run_python "${PYTHON_INTERNVL}" "${CONDA_INTERNVL}" \
        "${EVAL_DEPS_ROOT}/vlm/internvl_eval.py" \
        --video_dir "${VIDEO_PATH}" \
        --json_path "${INTERNVL_JSON}" \
        --output_dir "${INTERNVL_OUT}" \
        --summary_path "${INTERNVL_OUT}/summary.json" \
        --model_path "${INTERNVL_MODEL_PATH}" \
        --total_num "${INTERNVL_NUM_PROCS}" \
        --cur_num "${_cur}" \
        --no_summary
    ) &
  done
  wait
  CUDA_VISIBLE_DEVICES=0 _run_python "${PYTHON_INTERNVL}" "${CONDA_INTERNVL}" \
    "${EVAL_DEPS_ROOT}/vlm/internvl_eval.py" \
    --summary_only \
    --json_path "${INTERNVL_JSON}" \
    --output_dir "${INTERNVL_OUT}" \
    --summary_path "${INTERNVL_OUT}/summary.json" \
    --model_path "${INTERNVL_MODEL_PATH}"
  echo "[run_all] InternVL -> ${INTERNVL_OUT}/summary.json"
else
  echo "[run_all] SKIP InternVL"
fi

echo "[run_all] Done. Results: ${OUTPUT_DIR}"

	#!/bin/bash
	# Run all HOI evaluation metrics (self-contained paths under evaluation/).
	#
	# Prefer: bash scripts/run_eval.sh <benchmark> <method>
	# Or set VIDEO_PATH + source eval_common.sh, then run this script.
	#
	# Requires: bash scripts/setup_deps.sh /path/to/evaluations (once)
	# Optional: env.local.sh for PYTHON_* / model weights

	set -e

	SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
	# shellcheck source=eval_common.sh
	source "${SCRIPT_DIR}/eval_common.sh"

	if [ -z "${EVAL_DEPS_ROOT:-}" ] \|\| [ ! -d "${EVAL_DEPS_ROOT}/VBench" ]; then
	echo "[run_all] ERROR: metric code not found." >&2
	echo " Run: bash scripts/setup_deps.sh /path/to/projects/evaluations" >&2
	echo " Or: export EVAL_DEPS_ROOT=/path/to/evaluations" >&2
	exit 1
	fi

	VIDEO_PATH="${VIDEO_PATH:?Set VIDEO_PATH or use: bash scripts/run_eval.sh <bench> <method>}"
	OUTPUT_DIR="${OUTPUT_DIR:-${RESULTS_DIR}/$(basename "${VIDEO_PATH}")_eval}"

	OBJ_DINO_JSON="${OBJ_DINO_JSON:-${MANIFESTS_DIR}/obj_dino_crosstest.json}"
	TVA_JSON="${TVA_JSON:-${MANIFESTS_DIR}/tva_crosstest.json}"
	INTERNVL_JSON="${INTERNVL_JSON:-${DATASETS_PORTABLE}/hoi_crosstest_20251107.json}"

	SKIP_OBJ_DINO="${SKIP_OBJ_DINO:-0}"
	SKIP_VBENCH="${SKIP_VBENCH:-0}"
	SKIP_TVA="${SKIP_TVA:-0}"
	SKIP_INTERNVL="${SKIP_INTERNVL:-0}"

	PYTHON_OBJ_DINO="${PYTHON_OBJ_DINO:-}"
	CONDA_OBJ_DINO="${CONDA_OBJ_DINO:-dino_obj}"
	PYTHON_VBENCH="${PYTHON_VBENCH:-}"
	CONDA_VBENCH="${CONDA_VBENCH:-dl-main}"
	PYTHON_TVA="${PYTHON_TVA:-}"
	CONDA_VIDEOREWARD="${CONDA_VIDEOREWARD:-VideoReward}"
	PYTHON_INTERNVL="${PYTHON_INTERNVL:-}"
	CONDA_INTERNVL="${CONDA_INTERNVL:-internvl}"

	TVA_MODEL_PATH="${TVA_MODEL_PATH:-/path/to/VideoReward}"
	INTERNVL_MODEL_PATH="${INTERNVL_MODEL_PATH:-/path/to/InternVL3_5-38B}"
	TVA_NUM_GPUS="${TVA_NUM_GPUS:-8}"
	export VBENCH_CACHE_DIR="${VBENCH_CACHE_DIR:-${EVAL_DEPS_ROOT}/VBench/weghts}"

	mkdir -p "${OUTPUT_DIR}"

	# InternVL needs absolute ref-image paths when using datasets_portable/
	INTERNVL_JSON_RESOLVED="${OUTPUT_DIR}/.internvl_dataset.json"
	python3 "${SCRIPT_DIR}/absolutize_dataset_json.py" \
	--eval-root "${EVAL_ROOT}" \
	--input "${INTERNVL_JSON}" \
	--output "${INTERNVL_JSON_RESOLVED}"
	INTERNVL_JSON="${INTERNVL_JSON_RESOLVED}"

	_run_python() {
	local py="$1" conda_env="$2"
	shift 2
	if [ -n "${py}" ]; then
	"${py}" "$@"
	else
	conda run -n "${conda_env}" --no-capture-output --live-stream python "$@"
	fi
	}

	echo "[run_all] EVAL_ROOT=${EVAL_ROOT}"
	echo "[run_all] EVAL_DEPS_ROOT=${EVAL_DEPS_ROOT}"
	echo "[run_all] VIDEO_PATH=${VIDEO_PATH}"
	echo "[run_all] OUTPUT_DIR=${OUTPUT_DIR}"

	if [ "${SKIP_OBJ_DINO}" = "0" ]; then
	echo "[run_all] ----- 1/4 Object-DINO -----"
	OBJ_DINO_SAVE="${OUTPUT_DIR}/obj_dino_result.json"
	_run_python "${PYTHON_OBJ_DINO}" "${CONDA_OBJ_DINO}" \
	"${EVAL_DEPS_ROOT}/obj_dino/obj_dino.py" \
	--video_path "${VIDEO_PATH}" \
	--json_list "${OBJ_DINO_JSON}" \
	--save_path "${OBJ_DINO_SAVE}"
	echo "[run_all] Object-DINO -> ${OBJ_DINO_SAVE}"
	else
	echo "[run_all] SKIP Object-DINO"
	fi

	if [ "${SKIP_VBENCH}" = "0" ]; then
	echo "[run_all] ----- 2/4 VBench -----"
	VBENCH_OUT="${OUTPUT_DIR}/vbench_results"
	if [ -n "${PYTHON_VBENCH}" ]; then
	(cd "${EVAL_DEPS_ROOT}/VBench" && "${PYTHON_VBENCH}" evaluate.py \
	--videos_path "${VIDEO_PATH}" \
	--output_path "${VBENCH_OUT}" \
	--mode custom_input \
	--dimension motion_smoothness dynamic_degree aesthetic_quality overall_consistency)
	else
	conda run -n "${CONDA_VBENCH}" --no-capture-output --live-stream \
	bash -c "cd ${EVAL_DEPS_ROOT}/VBench && python evaluate.py \
	--videos_path \"${VIDEO_PATH}\" \
	--output_path \"${VBENCH_OUT}\" \
	--mode custom_input \
	--dimension motion_smoothness dynamic_degree aesthetic_quality overall_consistency"
	fi
	echo "[run_all] VBench -> ${VBENCH_OUT}"
	else
	echo "[run_all] SKIP VBench"
	fi

	if [ "${SKIP_TVA}" = "0" ]; then
	echo "[run_all] ----- 3/4 TVA (VideoAlign) -----"
	TVA_SAVE="${OUTPUT_DIR}/TVA_result.json"
	if [ -f "${EVAL_DEPS_ROOT}/VideoAlign/TVA_metrics_hzy_v2.py" ]; then
	if [ -n "${PYTHON_TVA}" ]; then
	(cd "${EVAL_DEPS_ROOT}/VideoAlign" && "${PYTHON_TVA}" TVA_metrics_hzy_v2.py \
	--video_path "${VIDEO_PATH}" \
	--json_list "${TVA_JSON}" \
	--save_path "${TVA_SAVE}" \
	--gpu_id 0)
	else
	conda run -n "${CONDA_VIDEOREWARD}" --no-capture-output --live-stream \
	bash -c "cd ${EVAL_DEPS_ROOT}/VideoAlign && python TVA_metrics_hzy_v2.py \
	--video_path \"${VIDEO_PATH}\" \
	--json_list \"${TVA_JSON}\" \
	--save_path \"${TVA_SAVE}\" \
	--gpu_id 0"
	fi
	else
	_run_python "${PYTHON_TVA}" "${CONDA_VIDEOREWARD}" \
	"${EVAL_DEPS_ROOT}/VideoAlign/TVA_metrics.py" \
	--model_path "${TVA_MODEL_PATH}" \
	--video_path "${VIDEO_PATH}" \
	--json_list "${TVA_JSON}" \
	--save_path "${TVA_SAVE}" \
	--num_gpus "${TVA_NUM_GPUS}"
	fi
	echo "[run_all] TVA -> ${TVA_SAVE}"
	else
	echo "[run_all] SKIP TVA"
	fi

	if [ "${SKIP_INTERNVL}" = "0" ]; then
	echo "[run_all] ----- 4/4 InternVL -----"
	INTERNVL_OUT="${OUTPUT_DIR}/internvl"
	INTERNVL_NUM_PROCS=8
	for _gpu in 0 1 2 3 4 5 6 7; do
	_cur="${_gpu}"
	(
	export CUDA_VISIBLE_DEVICES="${_gpu}"
	_run_python "${PYTHON_INTERNVL}" "${CONDA_INTERNVL}" \
	"${EVAL_DEPS_ROOT}/vlm/internvl_eval.py" \
	--video_dir "${VIDEO_PATH}" \
	--json_path "${INTERNVL_JSON}" \
	--output_dir "${INTERNVL_OUT}" \
	--summary_path "${INTERNVL_OUT}/summary.json" \
	--model_path "${INTERNVL_MODEL_PATH}" \
	--total_num "${INTERNVL_NUM_PROCS}" \
	--cur_num "${_cur}" \
	--no_summary
	) &
	done
	wait
	CUDA_VISIBLE_DEVICES=0 _run_python "${PYTHON_INTERNVL}" "${CONDA_INTERNVL}" \
	"${EVAL_DEPS_ROOT}/vlm/internvl_eval.py" \
	--summary_only \
	--json_path "${INTERNVL_JSON}" \
	--output_dir "${INTERNVL_OUT}" \
	--summary_path "${INTERNVL_OUT}/summary.json" \
	--model_path "${INTERNVL_MODEL_PATH}"
	echo "[run_all] InternVL -> ${INTERNVL_OUT}/summary.json"
	else
	echo "[run_all] SKIP InternVL"
	fi

	echo "[run_all] Done. Results: ${OUTPUT_DIR}"