open-mmlab / mmdetection3d

OpenMMLab's next-generation platform for general 3D object detection.
https://mmdetection3d.readthedocs.io/en/latest/
Apache License 2.0
5k stars 1.49k forks source link

[Bug] bevfusion image feature not working #2951

Open hipforth opened 2 months ago

hipforth commented 2 months ago

Prerequisite

Task

I'm using the official example scripts/configs for the officially supported tasks/models/datasets.

Branch

main branch https://github.com/open-mmlab/mmdetection3d

Environment

bevfusion

Reproduces the problem - code sample

mmdet3d code

Reproduces the problem - command or script

normal run

Reproduces the problem - error message

as follows

Additional information

I tested lidar+CAM_FRONT, lidar+CAM_BACK, and the detection results were exactly the same. Then I set the input image to pure black, and the detection results were still normal. But when the lidar input is set to an empty point cloud, the detection result is a bunch of chaotic BBoxes and the scores are all less than 0.1

zhangshen12356 commented 1 month ago

Prerequisite

Task

I'm using the official example scripts/configs for the officially supported tasks/models/datasets.

Branch

main branch https://github.com/open-mmlab/mmdetection3d

Environment

bevfusion

Reproduces the problem - code sample

mmdet3d code

Reproduces the problem - command or script

normal run

Reproduces the problem - error message

as follows

Additional information

I tested lidar+CAM_FRONT, lidar+CAM_BACK, and the detection results were exactly the same. Then I set the input image to pure black, and the detection results were still normal. But when the lidar input is set to an empty point cloud, the detection result is a bunch of chaotic BBoxes and the scores are all less than 0.1

  • lidar+black_image image
  • empty lidar + all rgb image image

If you find something about it and share what you get, I'll appreciate it !