Does RaCo reranker show clear advantage over score-based top-k selection?

[Vincentqyw]

Hi, thanks for the great work on RaCo!

I have a question regarding the practical benefit of the reranker when used in a detect → rerank → match pipeline.

Have you conducted experiments comparing reranker-based top-k vs detector-score-based top-k when feeding into a downstream matcher (e.g., LightGlue)?