imperfect surface reconstruction result on autonomous driving data

[LevinJ]

Hi, thanks for the great work. I tried applying 2d gs to road reconstruction, the overall result looks okay, but there are also obvious problems.

Below is the scene I try to reconstruct. it's about 20 meters long, and I collected 88 images by one single straight drive on the road.

The reconstructed result is as below,

There are at least three obvious defects, 1. holes on the road, 2. the sky points are too low. 3. the wall is distorted.

Is this the expected surface reconstruction result by 2dgs? What can we do to improve the reconstruction? Thanks.