scripts/aggregate_results.py

#!/usr/bin/env python3
"""Aggregate metric outputs into a single Excel summary.

Walks through one or more `*_eval` directories produced by
`run_all_metrics.sh` and produces a flattened summary table (one row per
method) covering Object-DINO / VBench / TVA / InternVL.

Usage:
  python aggregate_results.py --input_dir /path/to/<method>_eval
  python aggregate_results.py --input_dir /path/to/results --output summary.xlsx
  python aggregate_results.py --input_dirs /path/to/dir1 /path/to/dir2
"""
from __future__ import annotations

import argparse
import json
import os
import glob


def _read_json(path: str) -> dict | list | None:
    if not path or not os.path.isfile(path):
        return None
    try:
        with open(path, "r", encoding="utf-8") as f:
            return json.load(f)
    except Exception:
        return None


def collect_obj_dino(root: str) -> dict:
    p = os.path.join(root, "obj_dino_result.json")
    data = _read_json(p)
    if not data or not isinstance(data, dict):
        return {}
    return {
        "obj_dino_avg_dino_subject": data.get("avg_dino_subject"),
        "obj_dino_total_frames": data.get("total_frame_count"),
    }


def collect_vbench(root: str) -> dict:
    out = {}
    vbench_dir = os.path.join(root, "vbench_results")
    if not os.path.isdir(vbench_dir):
        return out
    candidates = glob.glob(os.path.join(vbench_dir, "*_eval_results.json"))
    if not candidates:
        return out
    latest = max(candidates, key=os.path.getmtime)
    data = _read_json(latest)
    if not data or not isinstance(data, dict):
        return out
    for dim in ["motion_smoothness", "dynamic_degree", "aesthetic_quality", "overall_consistency"]:
        val = data.get(dim)
        if val is not None:
            if isinstance(val, (list, tuple)) and len(val) > 0:
                out[f"vbench_{dim}"] = val[0]
            else:
                out[f"vbench_{dim}"] = val
    return out


def collect_tva(root: str) -> dict:
    p = os.path.join(root, "TVA_result.json")
    data = _read_json(p)
    if not data or not isinstance(data, dict):
        return {}
    return {
        "tva_VQ": data.get("average_VQ_score"),
        "tva_MQ": data.get("average_MQ_score"),
        "tva_TA": data.get("average_TA_score"),
        "tva_Overall": data.get("average_Overall_score"),
        "tva_videos": data.get("total_videos_processed"),
    }


def collect_internvl(root: str) -> dict:
    p = os.path.join(root, "internvl", "summary.json")
    data = _read_json(p)
    if not data or not isinstance(data, dict):
        return {}
    avg = data.get("average_scores") or {}
    return {
        "internvl_object": avg.get("object_score"),
        "internvl_human": avg.get("human_score"),
        "internvl_interaction": avg.get("interaction_score"),
        "internvl_total_avg": avg.get("total_average_score"),
        "internvl_videos": data.get("successfully_evaluated"),
    }


def collect_one_dir(root: str, method_name: str | None = None) -> tuple[str, dict]:
    """Collect all metrics from a single eval directory.

    Returns (method_name, flat_metrics_dict).
    """
    name = method_name or os.path.basename(root.rstrip("/"))
    if name.endswith("_eval"):
        name = name[: -len("_eval")]
    row = {"method": name}
    row.update(collect_obj_dino(root))
    row.update(collect_vbench(root))
    row.update(collect_tva(root))
    row.update(collect_internvl(root))
    return name, row


def flatten_order() -> list[str]:
    """Column order for the summary table."""
    return [
        "method",
        "obj_dino_avg_dino_subject",
        "obj_dino_total_frames",
        "vbench_motion_smoothness",
        "vbench_dynamic_degree",
        "vbench_aesthetic_quality",
        "vbench_overall_consistency",
        "tva_VQ",
        "tva_MQ",
        "tva_TA",
        "tva_Overall",
        "tva_videos",
        "internvl_object",
        "internvl_human",
        "internvl_interaction",
        "internvl_total_avg",
        "internvl_videos",
    ]


def main():
    parser = argparse.ArgumentParser(description="Aggregate evaluation metrics into an Excel summary.")
    parser.add_argument(
        "--input_dir",
        type=str,
        default=None,
        help="Single eval directory, or a parent dir (will scan for *_eval children).",
    )
    parser.add_argument(
        "--input_dirs",
        type=str,
        nargs="+",
        default=None,
        help="Multiple eval directories.",
    )
    parser.add_argument(
        "--output",
        type=str,
        default="summary.xlsx",
        help="Output Excel path.",
    )
    args = parser.parse_args()

    dirs = []
    if args.input_dirs:
        dirs = [os.path.abspath(d) for d in args.input_dirs]
    elif args.input_dir:
        d = os.path.abspath(args.input_dir)
        if os.path.isdir(d):
            base = os.path.basename(d.rstrip("/"))
            if base.endswith("_eval"):
                dirs = [d]
            else:
                pattern = os.path.join(d, "*_eval")
                dirs = sorted(glob.glob(pattern))
                if not dirs:
                    dirs = [d]
    if not dirs:
        print("No eval directories specified or found.")
        return

    rows = []
    for d in dirs:
        _, row = collect_one_dir(d)
        rows.append(row)

    try:
        import pandas as pd
    except ImportError:
        print("Needs pandas and openpyxl: pip install pandas openpyxl")
        return

    order = flatten_order()
    all_keys = set()
    for r in rows:
        all_keys.update(r.keys())
    cols = [c for c in order if c in all_keys] + sorted(all_keys - set(order))
    df = pd.DataFrame(rows, columns=cols)

    with pd.ExcelWriter(args.output, engine="openpyxl") as w:
        df.to_excel(w, sheet_name="summary", index=False)

    print(f"Written: {args.output}")
    print(f"Total methods: {len(rows)}")


if __name__ == "__main__":
    main()