NVIDIA-ISAAC-ROS
Using OpenCV NMS in decoder
I was using detectnet in ROS2 foxy but I want to update to the latest release in Humble. Before I was not using the DBscan clustering on the bounding boxes but I implemented opencv NMS in the decoder which gave me a bit better results and was faster. Now I want to do something similar but it done somewhere in the gxf library which is more difficult to change.
So my question is if I can use DetectNet with only NMS somewhat and no DBscan.
We'll add these options in an upcoming release but you'll need to reimplement the DetectNet decoder again to remove DBScan. The code has been moved to the GXF library to take advantage of NITROS but you can take a look at the earlier EA3 version from March 2022 which still had the code inline.
