PRBonn / deep-point-map-compression

MIT License
95 stars 28 forks source link

some clusters in poiunt cloud after decoding #1

Closed MaybeOjbk closed 2 years ago

MaybeOjbk commented 3 years ago

I don't know if there are some mistakes in your code that the point cloud after decoding have some clusters , image image

MaybeOjbk commented 3 years ago

or I need to change parameters which can solve the problem

louis-wiesmann commented 3 years ago

Hello @MaybeOjbk, yes it can happen that point clusters occur. If this is unwanted i suggest either spatially filter them out (e.g. voxelgrid) or try to add an repulsive loss on the points of the output.

xbais commented 2 years ago

Hello @louis-wiesmann , do you have any idea as to why the clustering is happening? Can I talk to you on email?

louis-wiesmann commented 2 years ago

There is nothing which forbids or punishes the network for creating cluster (as mentioned a repulsive loss could do this). But this behavior is not necessarily undesired. In the upsampling process each point creates a couple of other points. The upsampling ratio was computed heuristically and is for each point the same. There is no mechanism to tell a point when to stop the upsampling, so it will create more points but at the same location.

louis-wiesmann commented 2 years ago

To get rid of the clusters I used simply grid based subsampling: from Open3d.