Function run

scripts/detect_batch.py:73–195

Run batch detection inference.

(cfg: dict, img1: np.ndarray, img2: np.ndarray)

def run(cfg: dict, img1: np.ndarray, img2: np.ndarray) -> bool:
    """Run batch detection inference."""
    name = cfg["name"]
    out = OUTPUT_DIR / name
    out.mkdir(parents=True, exist_ok=True)

    sep(name)
    print(f"  model     : {cfg['model_path']}")
    print(f"  threshold : {cfg['confidence_threshold']}")
    print(f"  image_size: {cfg['image_size']}")
    print(f"  expected  : {cfg['expected_cls']}")

    # ── load model ────────────────────────────────────────────
    print("\n  Loading model ...")
    try:
        model = AutoDetectionModel.from_pretrained(
            model_type="huggingface",
            model_path=cfg["model_path"],
            confidence_threshold=cfg["confidence_threshold"],
            image_size=cfg["image_size"],
            device="cpu",
        )
    except Exception as e:
        print(f"  SKIP — {e}")
        return True

    detected = "sigmoid" if model._uses_sigmoid_cls else "softmax"
    match = detected == cfg["expected_cls"]
    print(f"  num_labels: {model.num_categories}")
    print(f"  cls type  : {detected}  (expected: {cfg['expected_cls']})  [{'OK' if match else 'MISMATCH'}]")

    images = [img1, img2]
    image_names = [Path(IMAGE1).stem, Path(IMAGE2).stem]

    # ── single-image detection + save ─────────────────────────
    sep("detection results")
    single_counts = []
    for i, (img, img_name) in enumerate(zip(images, image_names)):
        model.perform_inference(img)
        model.convert_original_predictions()
        preds = model.object_prediction_list_per_image[0]
        single_counts.append(len(preds))

        visualize_object_predictions(
            image=img,
            object_prediction_list=preds,
            output_dir=str(out),
            file_name=f"single_{img_name}",
        )

        print(f"\n  [{img_name}]  {len(preds)} detections")
        print(f"  saved -> {out / f'single_{img_name}.png'}")
        print_top_preds(preds)

    # ── batch detection + save ────────────────────────────────
    model.perform_batch_inference(images)
    model.convert_original_predictions(
        shift_amount=[[0, 0]] * len(images),