search

charlesq34 / pointnet

PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation
Other
4.83k stars 1.46k forks source link

Semantic Segmentation of own data - Parts of cloud disappears? #201

Closed Naor92 closed 5 years ago

Naor92 commented 5 years ago

Hey guys, so I've successfully been able to segment a point cloud of my living room by training my model using the provided 3D indoor parsing dataset. Here is a picture of the old result:

2019-08-01

I don't know what could have possibly changed but now running batch_inference seems to cut out half the points of the cloud and not running the segmentation correctly. Like this:

Screenshot from 2019-09-26 12-45-58

Anyone have a clue what could be causing that behavior? I even reran the training and got decent accuracy results but no change when trying to segment my own point clouds.

Naor92 commented 5 years ago

For anyone that was running into a similar issue - I was exporting these as PLY files and then manually converting to text files. Seems this was causing the issue. Exporting straight to an XYZRGB file seemed to fix the problem for some strange reason.

I'm still having an issue with large amounts of points simply disappearing from the cloud after segmentation. Not sure whats's that about yet.

discretecoder commented 4 years ago

How did you infer? just with XYZ and not with XYZRGBL format?

Naor92 commented 4 years ago

I used just XYZRGB format. I was using Kinects not Lidar