Segmentation fault when loading semanticKITTI dataset

[fredpeertje]

I installed Point Labeler in Ubuntu 20.04 VMware. Because I got an openGL error invalid enumerant(1280), I had to comment lines with glow::_CheckGlError(__FILE__, __LINE__); in viewport.cpp. Now when I load folder 00 from the semanticKITTI dataset I get the following error:

OpenGL Context Version 4.1 core profile
GLEW initialized.
OpenGL context version: 4.1
OpenGL vendor string  : VMware, Inc.
OpenGL renderer string: SVGA3D; build: RELEASE;  LLVM;
temp mem size: 3.6 MB
Found 20 instanceable labels.
-- Setting 'tile size' to 100
mem size: 2400 MB
-- Setting 'max scans' to 500
-- Setting 'min range' to 2.5
-- Setting 'max range' to 50
/home/fred/catkin_ws/src/point_labeler/bin
#tiles  = 40
Reading instances.txt...finished.
615 point clouds read.
0, 0, 100
Setting points...Segmentation fault (core dumped)

I think it has to do with the folder structure. So this is what I did: I installed the poses files form this link: http://www.cvlibs.net/download.php?file=data_odometry_poses.zip and renamed 00.txt to poses.txt and moved it to the folder 00. I downloaded the labels and velodyne folders from http://www.semantic-kitti.org/dataset.html. And the calib.txt files http://www.cvlibs.net/download.php?file=data_odometry_calib.zip.

[jbehley]

It seems that the OpenGL driver has some problems. It might be better to use a native graphics driver provided by the vendor of the graphics card.

The folder structure should be the same as described here: https://github.com/jbehley/point_labeler#folder-structure

The pose files that we provide with the dataset are the ones that we used to annotate the data; these poses also include loop closures.

[fredpeertje]

So, the poses.txt files are not the ones provided in this zip? http://www.cvlibs.net/download.php?file=data_odometry_poses.zip Where can I download the poses.txt files? I tried the SuMA tool to create them but got the same issue as described here: https://github.com/jbehley/SuMa/issues/16#issue-437664598

[jbehley]

We used our Surfel-based Mapping approach to perform mapping and loop closures (some loop closures were manually tagged). Therefore, our poses will have consistent poses when revisiting areas, which are not provided by the GPS/INS poses.

The poses are included in our SemanticKitti release: http://semantic-kitti.org/dataset.html#download

Regarding the save dialog issue, I can unfortunately not reproduce the error. I never tested my software in VMWare and cannot provide any support for this.

[fredpeertje]

Ah the poses.txt files are in this zip file http://www.semantic-kitti.org/assets/data_odometry_labels.zip. Unfortunately same error, segmentation fault.

[jbehley]

As I said, it's unlikely that it's a problem with the files, but with the graphics driver.

It might be that I'm using somewhere a feature which is not required by the specification of the core profile of OpenGL. Therefore, it says "invalid enumerant". Nvidia drivers are less picky and silently accept also some non-core functions, texture types, etc. but finding this is really hard for me, since I use a Nvidia GPU with nvidia drivers.

Thus my suggestion to try with proper graphics drivers.

[jbehley]

Since there is no activity since some time, I close this issue now. If you have further comments on this issue, you can reopen it. Thanks.