search

LightwheelAI / street-gaussians-ns

Unofficial implementation of "Street Gaussians for Modeling Dynamic Urban Scenes"
Apache License 2.0
199 stars 18 forks source link

ERROR: Linear solver failure. Failed to compute a step: Eigen failure. Unable to perform dense Cholesky factorization. #9

Closed FudongGe closed 4 weeks ago

FudongGe commented 4 weeks ago

Hi @LightwheelAI ,

When I run "sh scripts/shells/run_colmap.sh $root", it happens that the linear solver failure, just like the picture below. Can I do something to solve this problem? Thanks in advance.

image

FudongGe commented 4 weeks ago

And maybe there is something wrong in the 'masks_generate.py'? The masks obtained from the 'masks_generate.py' and the masks you provide are so different, just like the pictures below (They correspond to the same image). I have no idea what makes it happen. Do you have any ideas?

1

2

LightwheelAI commented 4 weeks ago

Thanks for your trying! Sry, we can't solve the problem you're having with colmap for now. Regarding masks, our masks pick instance masks from the segment solution based on the 2D box rendered by the bbox, whereas here it's the 2D box rendered by the bbox.The main role of mask is to prevent the presence of dynamic target floaters in the SfM point cloud generated by colmap, so the result won't be too different in experiments.

FudongGe commented 4 weeks ago

Thanks for your trying! Sry, we can't solve the problem you're having with colmap for now. Regarding masks, our masks pick instance masks from the segment solution based on the 2D box rendered by the bbox, whereas here it's the 2D box rendered by the bbox.The main role of mask is to prevent the presence of dynamic target floaters in the SfM point cloud generated by colmap, so the result won't be too different in experiments.

Thanks for your trying! Sry, we can't solve the problem you're having with colmap for now. Regarding masks, our masks pick instance masks from the segment solution based on the 2D box rendered by the bbox, whereas here it's the 2D box rendered by the bbox.The main role of mask is to prevent the presence of dynamic target floaters in the SfM point cloud generated by colmap, so the result won't be too different in experiments.

Ok, thanks for your response. And I will try to solve the problem.