UIKA / tools /metrics /evaluate.py
vegagoto123's picture
Add files using upload-large-folder tool
25a2c48 verified
Raw History Blame Contribute Delete
2.64 kB
from __future__ import annotations
import argparse
import sys
from pathlib import Path
REPO_ROOT = Path(__file__).resolve().parents[2]
if str(REPO_ROOT) not in sys.path:
sys.path.insert(0, str(REPO_ROOT))
DEFAULT_DEEP3D_CHECKPOINT = "model_zoo/tools/deep3dface_recon_2023ver_epoch_20.pth"
def _print_dependency_notes(mode: str, deep3d_checkpoint: str | Path, insightface_name: str) -> None:
print("Metric dependencies:")
print(f"- AED/APD: {Path(deep3d_checkpoint).expanduser()}")
print(f"- CSIM: InsightFace `{insightface_name}` cache under ~/.insightface/models")
if mode == "self":
print("- AKD: face_alignment cache under ~/.cache/torch/hub/checkpoints")
print("- LPIPS: lpips package weights plus torchvision AlexNet cache when needed")
def parse_args() -> argparse.Namespace:
parser = argparse.ArgumentParser(description="Compute UIKA image quality and face metrics.")
parser.add_argument("--pred-dir", required=True, help="Directory containing predicted frames")
parser.add_argument("--gt-dir", required=True, help="Directory containing ground-truth frames")
parser.add_argument("--output", required=True, help="Output CSV path")
parser.add_argument("--ref-image", default=None, help="Reference image for cross reenactment CSIM")
parser.add_argument("--mode", choices=["self", "cross"], default="self", help="Evaluation mode")
parser.add_argument("--device", default=None, help="Torch device, default: cuda:0 when available else cpu")
parser.add_argument(
"--deep3d-checkpoint",
default=DEFAULT_DEEP3D_CHECKPOINT,
help="Deep3DFaceRecon checkpoint path used for AED/APD",
)
parser.add_argument("--insightface-name", default="buffalo_l", help="InsightFace model pack name")
parser.add_argument(
"--on-error",
choices=["fail", "skip"],
default="fail",
help="Whether to fail immediately or skip frame pairs when one metric fails",
)
return parser.parse_args()
def main() -> None:
args = parse_args()
_print_dependency_notes(args.mode, args.deep3d_checkpoint, args.insightface_name)
from tools.metrics.calculator import MetricsCalculator
calculator = MetricsCalculator(
mode=args.mode,
device=args.device,
deep3d_checkpoint=args.deep3d_checkpoint,
insightface_name=args.insightface_name,
)
calculator.process_directories(
pred_dir=args.pred_dir,
gt_dir=args.gt_dir,
output_path=args.output,
ref_image=args.ref_image,
on_error=args.on_error,
)
if __name__ == "__main__":
main()