INNER CODE UNIT · Python

run_model

openfoodfacts/openfoodfacts-ai · logo-ann/benchmarks/embedding_models_benchmark/main.py:133

def run_model(
    root_dir: Path,
    split_set: Optional[Set[str]],
    model_name: str,
    batch_size: int,
    device: torch.device,
) -> Tuple[torch.Tensor, torch.Tensor, LogoDataset, float]:
    model = timm.create_model(model_name, pretrained=True, num_classes=0)
    model.to(device)
    config = resolve_data_config({}, model=model)
    transform_func = create_transform(**config)
    dataset = LogoDataset(
        root_dir, transform_func, split_set
    )  # our dataset is built on the data written in val.txt
    data_loader = DataLoader(dataset, batch_size, num_workers=2)
    embeddings, labels, elapsed = generate_embeddings(model, data_loader, device)
    return embeddings, labels, dataset, elapsed

View source record →