Function evaluate

tools/test_zero_shot_hf.py:401–441

Run zero-shot evaluation. Args: model: VTPModel instance. classifier: Zero-shot classifier weights, shape (embed_dim, num_classes). dataloader: ImageNet validation dataloader. device: Device to use. precision: Precision for inference. Returns:

(
    model: VTPModel,
    classifier: torch.Tensor,
    dataloader: DataLoader,
    device: torch.device,
    precision: str = 'fp32',
)

# ============================================================================

def evaluate(
    model: VTPModel,
    classifier: torch.Tensor,
    dataloader: DataLoader,
    device: torch.device,
    precision: str = 'fp32',
) -> Tuple[float, float]:
    """Run zero-shot evaluation.

    Args:
        model: VTPModel instance.
        classifier: Zero-shot classifier weights, shape (embed_dim, num_classes).
        dataloader: ImageNet validation dataloader.
        device: Device to use.
        precision: Precision for inference.

    Returns:
        Tuple of (top1_accuracy, top5_accuracy).
    """
    autocast_ctx = get_autocast_context(precision, device_type=device.type)
    input_dtype = get_input_dtype(precision)

    top1, top5, n = 0.0, 0.0, 0

    with torch.inference_mode():
        for images, targets in tqdm(dataloader, desc="Evaluating"):
            images = images.to(device=device, dtype=input_dtype)
            targets = targets.to(device)

            with autocast_ctx():
                image_features = model.get_clip_image_feature(images, normalize=True)
                logits = 100.0 * image_features @ classifier

            acc1, acc5 = accuracy(logits, targets, topk=(1, 5))
            top1 += acc1
            top5 += acc5
            n += images.size(0)

    top1 = top1 / n * 100
    top5 = top5 / n * 100
    return top1, top5


# ============================================================================