Scaffold-GS metrics only uses 7 scenes.

[jefequien]

The Scaffold-GS PSNR of 28.84 is across 7 scenes, not 9 as used by the rest of the table. It should be 27.50 as reported by the HAC paper which uses all 9 scenes.

https://github.com/w-m/3dgs-compression-survey/blob/5b1b110a899c44482c2ace0758bcdfcfeba5a4bc/results/MipNeRF360.csv#L12

Screenshots from Scaffold-GS paper.

Screenshots from HAC paper.

[w-m]

Thanks, Jeffrey, for the report!

@paulbzm then please use the numbers from the HAC paper for Scaffold-GS in the table

@myla maybe highlight this point in the author mailing to the Scaffold-GS authors

[jefequien]

@w-m For your awareness, I also opened this issue. https://github.com/YihangChen-ee/HAC/issues/14

[paulbzm]

Thanks @jefequien , the results are updated now!

[jefequien]

@paulbzm The HAC authors responded to my issue. You can use the new HAC numbers in the survey. Also, the Scaffold-GS numbers are likely incorrect because they do not consider MipNeRF's data factor.

[paulbzm]

Thanks again @jefequien !