def load(self, device: torch.device) -> None
Materialize weights/processor on ``device`` (call once before use).
Source: unitrack/benchmarks/hota/protocols.py:23
