Transfered To Nuscenes Dataset，Performance decline about lidar_rcnn HOT 10 CLOSED

The nuscenes dataset has a multi-frame setting becasue of the 20HZ 32L LiDAR. Do you add the delta time in input points? @Suodislie

from lidar_rcnn.

Thank you very much for your reminder, I just added a keyframe point cloud. Have you reproduced this idea in nuscense? In addition, what are the more important steps different from waymo, such as the selection of positive and negative samples, expand_proposal_meter parameters, etc.

from lidar_rcnn.

@Suodislie
I haven't try it in nuscense. Maybe you should pay attention to positive and negative samples.

from lidar_rcnn.

Ok. Ths for your suggestions ！

from lidar_rcnn.

@Suodislie If there are some new results in nuscenes please let me know. Feel free to contact me for further discussions!

from lidar_rcnn.

Ok. Ths for your suggestions ！

hallo！can you share the result of lidar-rcnn in nuscenes? Performance decline exactly?

from lidar_rcnn.

@Lzc6996 hi~can you share your experiment on nuscenes?

from lidar_rcnn.

@hutao568 @xxlbigbrother
Sorry, I didn't conduct a experiment on nuscenes.
For 10 frames point cloud as input，there are so many points not fall in the proposal box. Speed prediction can help us to find these points we need. But I haven't do it on nuscenes dataset.

from lidar_rcnn.

Have figure out this problem? I trans this code to nuscnes on pointpillar, which lead to 10 percent decline, @Suodislie

from lidar_rcnn.

@hutao568 @xxlbigbrother Sorry, I didn't conduct a experiment on nuscenes. For 10 frames point cloud as input，there are so many points not fall in the proposal box. Speed prediction can help us to find these points we need. But I haven't do it on nuscenes dataset.

Will this algorithm perform worse in scenarios with more detailed categories (such as nuscenes)?

I'm not sure if you have tested on other data with more categories from waymo and kitti. Looking forward to your reply!

from lidar_rcnn.