search

hugoycj / 2.5d-gaussian-splatting

an unofficial 2DGS implementation based on GauStudio
https://github.com/GAP-LAB-CUHK-SZ/gaustudio
Other
220 stars 4 forks source link

Confused about the results #3

Closed lif314 closed 2 months ago

lif314 commented 2 months ago

I tested on Tandt/Truck, and the results were not satisfactory. the 'gaussian' is not a 'disk' like in 2DGS. It has values for all three scales.

Gaussian Balls: image

Mesh: image

hugoycj commented 2 months ago

That's an interesting issue with the visualization. We've checked the scaling of the trained model and confirmed that the third axes are all 0. That suggests the problem may lie in the visualization process itself.

image

Maybe we need to dig into SIBR_viewer source code to identify any issues or inconsistencies in how the visualization is being generated.

lif314 commented 2 months ago

That's an interesting issue with the visualization. We've checked the scaling of the trained model and confirmed that the third axes are all 0. That suggests the problem may lie in the visualization process itself.

image

Maybe we need to dig into SIBR_viewer source code to identify any issues or inconsistencies in how the visualization is being generated. There is nothing wrong with your code. It was my carelessness that caused this issue. I use GaussianSplattingView to visualize Gaussians, this problem should be a bug of GaussianSplattingView. I'm sorry for wasting your time with this issue! Thanks for your great work, I will close this issue.

image

city19992 commented 1 month ago

I noticed that the figures in comments are can not be seen. Is there anything wrong?