fp_test.py

"""Quick false-positive check: run inference on the test split and flag any
frames where confidences are suspiciously low (likely FP-prone range).

Usage:  python fp_test.py [--conf 0.35] [--weights path/to/best.pt]
"""

import argparse
from pathlib import Path
from ultralytics import YOLO


def main():
    parser = argparse.ArgumentParser(description="False-positive diagnostic tool")
    parser.add_argument(
        "--weights",
        default="runs/detect/coin_detector_nano_v2/weights/best.pt",
        help="Path to model weights",
    )
    parser.add_argument(
        "--conf", type=float, default=0.35, help="Confidence threshold"
    )
    parser.add_argument(
        "--source",
        default="datasets/images/test",
        help="Image folder to scan",
    )
    args = parser.parse_args()

    model = YOLO(args.weights)
    source = Path(args.source)
    if not source.exists():
        print(f"Source folder not found: {source}")
        return

    images = sorted(
        p for p in source.iterdir() if p.suffix.lower() in {".jpg", ".jpeg", ".png"}
    )
    print(f"Scanning {len(images)} images at conf={args.conf}\n")

    low_conf_detections = []
    total_dets = 0

    for img in images:
        results = model(str(img), conf=args.conf, verbose=False)
        boxes = results[0].boxes
        total_dets += len(boxes)

        for box in boxes:
            c = float(box.conf[0])
            if c < 0.50:  # flag anything in the "shaky confidence" range
                cls_name = model.names[int(box.cls[0])]
                low_conf_detections.append((img.name, cls_name, c))

    print(f"Total detections: {total_dets}")
    if low_conf_detections:
        print(f"\n⚠  {len(low_conf_detections)} low-confidence detections (conf < 0.50):")
        for name, cls, c in low_conf_detections:
            print(f"  {name:40s}  {cls:10s}  conf={c:.3f}")
        print("\nThese are the most likely false positives. Inspect the images above.")
    else:
        print("\n✅ No low-confidence detections — looks clean.")


if __name__ == "__main__":
    main()