Significant difference in SampleBEV metrics between original paper and this repo

[caiusy]

I noticed that the SampleBEV metrics reported in the original paper are much higher than what I observe when running the code in this repo. Could you help clarify what causes this gap (e.g., differing settings, data splits, visibility filters, evaluation protocols)?

[caiusy]

Simple-BEV[13]RN-50 36.9∗ 40.9∗ 43.0∗ 46.6 in pointbev paper
Simple-BEV[7] RN-50 36.9 40.9 43.0 44.9 in your paper
Camera 47.4 in samplebev paper