tools/run_eval.py

"""Top-level evaluation script — runs Protocols 1/2/3/4 on a checkpoint."""
from __future__ import annotations

import argparse
import json
import sys
from pathlib import Path

import torch
from omegaconf import OmegaConf

sys.path.insert(0, str(Path(__file__).resolve().parent.parent))

from src.data import FairTalkingDataModule
from src.methods import build_method
from src.eval import run_protocol_1, run_protocol_2, run_protocol_3, run_protocol_4


def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("--config", required=True, help="path to the Hydra-composed config saved during training")
    ap.add_argument("--ckpt", required=True)
    ap.add_argument("--out_dir", required=True)
    ap.add_argument("--protocols", default="1,2,3,4",
                    help="comma-separated list of protocol IDs to run")
    args = ap.parse_args()

    cfg = OmegaConf.load(args.config)
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

    dm = FairTalkingDataModule(data_cfg=cfg.data, return_paired=False)

    model = build_method(
        method_name=cfg.method.name,
        method_cfg=cfg.method,
        backbone_cfg=cfg.backbone,
        data_cfg=cfg.data,
    )
    state = torch.load(args.ckpt, map_location="cpu")
    model.load_state_dict(state.get("state_dict", state), strict=False)

    out = Path(args.out_dir); out.mkdir(parents=True, exist_ok=True)

    prots = [p.strip() for p in args.protocols.split(",") if p.strip()]
    all_metrics = {}
    if "1" in prots:
        all_metrics["protocol1"] = run_protocol_1(model, dm, device, out)
    if "2" in prots:
        all_metrics["protocol2"] = run_protocol_2(model, dm, device, out)
    if "3" in prots:
        all_metrics["protocol3"] = run_protocol_3(model, dm, device, out)
    if "4" in prots:
        all_metrics["protocol4"] = run_protocol_4(model, dm, device, out)

    (out / "summary.json").write_text(json.dumps(all_metrics, indent=2))
    print(json.dumps(all_metrics, indent=2))


if __name__ == "__main__":
    main()