VD-Vidore-V2 evaluation metrics do not match the paper.

[Lebron-James777]

When testing VLM2VEC-V2, we found significant differences between the metrics of the four datasets in VD-Vidore-V2 and those in the paper, while the metrics of other datasets were basically consistent. Is any special setting required?

[SkyFishMoon]

I met the same problem.

[SkyFishMoon]

just note that the metrics for the visdoc subsets is NDCG@5 rather than Hit@1. Make sure you are comparing the right metrics.