Reproduction Discrepancy on Lerf Dataset

[sohcahtoae]

Hi, thanks for releasing this great project! I’m trying to reproduce the results from your paper on the Lerf dataset.

For the teatime dataset, I obtained the following results:

Mean IoU per class: {'sheep': 0.5264876460419082, 'plate': 0.449207029082698, 'apple': 0.9236299767545247}
Mean Boundary IoU per class: {'sheep': 0.4352036218062042, 'plate': 0.4511066046043869, 'apple': 0.9207977168668688}
Overall Mean IoU: 0.6331082172930437
Overall Boundary Mean IoU: 0.60236931442582

However, the results are quite lower than the MIoU (81.2) stated in the paper.
Could you provide some insight into what might be causing this discrepancy?

[Frederick-z]

This paper appears to be a simple combination of LangSplat and Gaussian Grouping, so why does it significantly improve performance?