hbb1/2d-gaussian-splatting

A question about specular highlight region‘s reconstruction.

Opened this issue · 1 comments

Hi hbb1,
I have got wonderful results using 2DGS, Thanks for your amazing work!

As mentioned in the paper and some issues, specular highlight region is hard to reconstruct. But I check the .ply generated, I think the number of gaussian points in specular area is not sparse, then why would specular highlight area's mesh always end up with hollow holes?(我看了一下生成的ply文件,感觉高光区域的点的数量并不算少,但是为啥这些地方的重建效果可能会有空洞?)

I'd appreciate it if you could reply!

hbb1 commented

尽管有高光区域有足够多的点,但是也是容易产生空洞的。因为这些点在fake高光,产生的 depth 就会与实际的曲面不一致,导致空洞。