about dataset of segmentation and detection

[yu8ri1]

Is it possible to map data for segmentation and object detection?

[whyekit-motional]

@yu8ri1 pls check out the map expansion (tutorial can be found here: https://colab.research.google.com/github/nutonomy/nuscenes-devkit/blob/master/python-sdk/tutorials/map_expansion_tutorial.ipynb)

[yu8ri1]

I may have misled you with my poor English ... . My question is: Is the point cloud for semantic segmentation and the point cloud for 3D object detection the same point cloud used? If so, I would like to know how to know the correspondence.

[whyekit-motional]

@yu8ri1 yes, the point cloud for semantic segmentation and the point cloud for 3D object detection is exactly the same

In other words, you simply need to load nuScenes to get the lidar point cloud (xyz, intensity, etc.) and then load Panoptic nuScenes to get the semantic label for each lidar point (there is a one-to-one correspondence between the two)

For a concrete example on how to do this, pls see https://github.com/nutonomy/nuscenes-devkit/issues/650#issuecomment-907540001