martinakos

Basalt has released code for ICCV'21 paper "Square Root Marginalization for Sliding-Window Bundle Adjustment" in their master branch. This makes their code 36% faster while not losing any accuracy. https://vision.in.tum.de/research/vslam/rootvo...

Thanks for the patch @giampaolo I've used your patch to get the per thread cpu utilization and plot it with a modified psrecord. Below I show the psrecord output showing...

I also have this exception. Is there any way to avoid it starting from a stationary camera?

Maybe related to my issue #164

I see the same artifacts. Is there any explanation for these?

That showed the problem! I was writing to the images.txt the camera to world transformations and it should've been the other way round, world to camera. I corrected this and...

I suspect it may be related with yak depth image being unsigned short type while gazebo gazebo_ros_openni_kinect plugin producing a 32FC1 image.

I actually prefer to use the old basalt now, as it's faster. See this unanswered question about performance: https://gitlab.com/VladyslavUsenko/basalt/-/issues/90

I'd also like to know that

I'm already using --SiftExtraction.estimate_affine_shape 1 --SiftExtraction.domain_size_pooling 1 --SiftMatching.guided_matching 1 what else can I change?