weiyithu / SurroundOcc

[ICCV 2023] SurroundOcc: Multi-camera 3D Occupancy Prediction for Autonomous Driving
Apache License 2.0
821 stars 105 forks source link

Is it possible to create a panoptic occupancy map using your code? #24

Open joonsu0109gh opened 1 year ago

joonsu0109gh commented 1 year ago

I want to create a panoptic occupancy map using panoptic LiDAR labels. I'm curious if it's possible to generate a panoptic occupancy map using your code.

weiyithu commented 1 year ago

Hi, you mean the panoptic occupancy labels or using the panoptic images as inputs?

joonsu0109gh commented 1 year ago

I mean panoptic occupancy labels.

Thank you for the prompt response.

weiyithu commented 1 year ago

Yeah, our generated labels are panoptic occupancy labels, not monocular ones.

joonsu0109gh commented 1 year ago

Oh, what I want to ask is whether I can add instance labels to the occupancy using NuScenes panoptic LiDAR segmentation labels. I apologize for any confusion.

lqzhao commented 1 year ago

Yes, you can add instance labels to the foreground objects according to the known bounding boxes. Each bounding box has a unique 'object token', which can be the key for you to distinguish different objects with the same semantic labels.

joonsu0109gh commented 1 year ago

Do you mean that I can add instance labels with a bounding box without using LiDAR panoptic segmentation labels?

lqzhao commented 1 year ago

Hi, I'm not very familiar with "LiDAR panoptic segmentation labels" in nuScenes. If nuScenes already provides point-wise "LiDAR panoptic segmentation labels", you can use them to replace the LiDAR semantic segmentation labels in our occupancy generation pipeline.

The bounding boxes are used in our generation pipeline to crop movable objects from the scene. If nuScenes doesn't provide point-wise "LiDAR panoptic segmentation labels", you can use 'object token' to generate instance-level occupancy labels.