search

traveller59 / second.pytorch

SECOND for KITTI/NuScenes object detection
MIT License
1.72k stars 721 forks source link

why do points seems to be cut by half when use custom data? #124

Open 1453042287 opened 5 years ago

1453042287 commented 5 years ago

2019-03-30 10-53-10屏幕截图 2019-03-30 10-52-39屏幕截图 like the photos above, any suggestions would be appreciate! BTW, thanks for the work!

traveller59 commented 5 years ago

This is a bug, you can uncomment the line 167 ~ 173 in second.kittiviewer.backend.py. I am working on nuscenes dataset and new dataset api, and a web tool which only interact with dataset api to fit custom data.

1453042287 commented 5 years ago

thanks for the response, but it didn't work for me, the point cloud i used is generated from depth map instead of lidar, so maybe there is something different because if i use the lidar's data, it works fine, vertical coordinates is different maybe... and the number of the cut points is always change. like there is a section with a specific angle. :( 2019-03-31 13-36-00屏幕截图

daviduhm commented 5 years ago

Hello @1453042287 ,

It will be much easier to collect custom data (LiDAR point clouds, camera images, and ground truth labels) using LGSVL Simulator. You can see that the visualizer is correctly showing point clouds in bird eye view as well as in 3D space. LGSVL Simulator is an open-source simulator for Autonomous Vehicles like Carla. Please try using it and let me know if you need further assistance.

lgsvl_visualize_v5

1453042287 commented 5 years ago

@daviduhm it's very kind of you, i will try it later :)