naitiknakrani-eic

We have the similar issue with 16UC1 depth data coming from ToF. we can rectify it with non-accelerated ros image pipeline but not with Isaac_ros_image_pipeline. More image encoding support are...

Is there any logic difference in depth compute library function compared to open3d point cloud from depth function (https://www.open3d.org/docs/0.7.0/python_api/open3d.geometry.create_point_cloud_from_depth_image.html) ?

Hi @matlabbe Can you please explain more about your idea? We analyzed timing and cluster segmentation using radiusSearch is taking huge computation time. We ae trying to optimize this time...

No not yet. Just wanted to understand your idea, how unordered and ordered point cloud effects on segmentation computation time. When you say normal clustering are you referring pcl::EuclideanClusterExtraction ?

@matlabbe Thanks. This helps a lot. For our database, we are getting graph like this,  We have Map assembling time, but its very less hence invisible in the graph....

@matlabbe If Memory management is disabled, and actual processing time exceeds the Rtabmap/detectionrate, then what happen ? - Does it skip the node creation ? and hence the nodes created...

@matlabbe That's fantastic to know how these parameters really help in optimizing computation time. "the rate is limited either by Rtabmap/DetectionRate or actual maximum rate rtabmap can process based on...

Hi @matlabbe Thanks for your explanation and sorry for delay in revert back. We were understanding code and logic for NormalSegmentation. It was surprise to see that the Rtabmap works...

> To improve proximity detection time, if you don't care not keeping the raw laser scans, is to downsample the scans with `--Mem/LaserScanVoxelSize 0.05` and precompute the normals (`--Mem/LaserScanNormalK 20`)....

We benchmarked mapping for two devices AGX orin and Orin nano on the same recorded bagfile. Here we converted depth into pointcloud with zero ROI and 2 decimation form pointcloud_xyz...