Is it possible to use results run on nuscenes and nuimages together

#1

Hello,
Thanks for sharing your dataset.
I am trying to run tracking on nuscenes mini version using EagerMot. There they have 3d detection run on nuscenes and 2d detections run on nuimages and trying to track these using nuscenes scene information. Im stuck at understanding if it is possible to get any relation between nuimges and nuscenes results to project on nuscenes data.
It would be really great if you can help me understand this relation between nuimages and nuscenes.

Thanks,
Gayathri.

#2

Hi @Gayathri81, there is little to no correspondence / relation between the frames in nuImages and nuScenes

#3

Thanks for the quick clarification.

#4

To add to that, while none of the images are expected to be in both datasets, they are using the same taxonomy, come from the same cars and are collected in the same locations. So nuImages should be very suitable for training an image network that you could apply to nuScenes. This is what we did for PointPainting (https://arxiv.org/abs/1911.10150).

#5

Thanks @holger-motional for sharing the information, will go through this paper you have shared.
Is it possible to get the segmenatation in nuscenes ?

#6

Hi. There is no image semantic segmentation ground-truth in nuScenes. We have lidar semantic segmenation ground-truth though: https://www.nuscenes.org/nuscenes#lidarseg