Abstract: Highlights•We propose a simple method maximizing given information, even if overlooked before.•We improve the performance of baseline models on several datasets.•Our method get better localization performance only with minimal test overhead.
Loading