Automatic Ground Classification issue in version 2.0.0 and 2.0.1

[JesseG138]

 A little background... I am a land surveyor in Washington State, USA and I currently use a DJI M300RTK and the P1 camera and L1 LiDAR sensor to survey various types of topography. I have been using Metashape for over 6 years but for one reason or another never set up an account for the forums.

 I noticed an extreme difference in how the automatic ground classification is classifying Low Points(Noise). All points under any other points are being classified as Low Points(Noise), even points that are under trees and building overhangs. This causes lots of problems since we need the ground points to be classified as ground in these areas to reconstruct an accurate 3D model of the terrain. We have large trees on nearly every site we survey so we end up with huge portions of the point cloud classified as noise rather than ground and must then do hours of manual classification to reclassify the noise points into the ground class. The algorithm is also not classifying random points around the point cloud that in previous versions meet the classification tolerances. 

 The automatic "classify points" routine does not do this but is not tweakable and ends up wrongly classifying large portions of the point cloud leading to the need for more manual classification. I also noticed that changing the "Confidence" setting in the classify points dialogue no longer has any effect on the resulting classification. I tried a range from 0.01 to 1.0 and got identical results each time.

Version 1.8.5 does not have this problem with the classify ground points routine and works as it has for all the years past. I have attached screen shots of the ground classification results in version 2.0.0 and from version 1.8.5 for comparison as well as shots from version 2.0.0 showing the results of the "classify points" routine ran with 0.01 confidence and 1.0 confidence settings.

I am hoping there is a fix for this since the added lidar functionality of version 2 is very helpful when working with and combining photo/lidar datasets so I would hate to have to stay in 1.8.5.
Please advise if anyone else is seeing the same thing in the new versions. I have already tried clearing all my settings to defaults which had no effect so I am fairly certain that it has to do with the new classification routine that incorporates the return number for lidar point clouds.

[Paulo]

same here just did a ground classification on v. 2.0.1 and low points are wrongly classified (under roofs or over hangs, along walls). Definitely this did not occur with 1.8...

[JesseG138]

Thank you for verifying it is not an isolated problem. I am really hoping there is a quick fix for this since version 2.0 seems to handle large point clouds much better than 1.8.5 which was a very welcomed improvement when working with LiDAR and photogrammetry data in the same project.

[Alexey Pasumansky]

Thank you for reporting the issue, we are analyzing the problem now.

[JPG_Z]

same here just did a ground classification on v. 2.0.1 and low points are wrongly classified (under roofs or over hangs, along walls). Definitely this did not occur with 1.8...

I'm getting the same problem.

Lets hope they found a fix.

[JesseG138]

Thank you for reporting the issue, we are analyzing the problem now.

Thank you for letting us know Alexey. I am hoping you find a fix for this since version 1.8.5 does not handle large point clouds as well as 2.0.1 does and the added lidar functionality in 2.0 is very valuable to my workflow in combining lidar and photogrammetry derived point clouds. Please let me know if there is anything I can do to help(provide logs, data, etc.)