How to preprocess the scene dataset

[HaolinLiu97]

I try to use the preprocessing script boundary sample to compute distance field for the scannet scene, which should be quiet similar to the Gibson dataset. However, I found that some distance field is not correctly computed after visualization (Actually most of the distance field is not computed corrected, for example, query points behind the walls will have weirdly very small distance field). I assume this is caused by reason that libigl is not able to compute the signed distance field on open surface. Therefore, I wonder how the author preprocess the scene data, will be preprocessing script for scene data released in the future? Thank you in advances!

[ZhengdiYu]

Could you share some of your visualization?

I have also tried to train on a single frame of ScanNet. And The reconstruction results seemed not bad. How did you inspect if the dfs are correct or not?

[HaolinLiu97]

@jchibane Sorry for the late reply, but I eventually figure it out it would be more suitable to compute the distance field using igl.point_mesh_squared_distance for open mesh. The sign_distance function still seems to have some problem with open mesh, but I guess it will not affect the training very much for small deviation from the actual distance field.

[jchibane]

Hey, yes, you need to compute unsigned distance fields - the sign you get can not be reliable in case of open surfaces. Maybe having a look into the pull request would be helpful to you.

Best, Julian

[ZhengdiYu]

@jchibane Sorry for the late reply, but I eventually figure it out it would be more suitable to compute the distance field using igl.point_mesh_squared_distance for open mesh. The sign_distance function still seems to have some problem with open mesh, but I guess it will not affect the training very much for small deviation from the actual distance field.

Thank you for your reply, but why is it available for cars' open surface?