[New Models] bevfusion

[betterhalfwzm]

Model/Dataset/Scheduler description

bevfusion 利用multi_modality_demo.py测试demo/nuscenes的例子，如果将图片给假的输入，保留雷达为真，结果基本正确，但是将雷达给一个假的输入，图片为真，则没有输出结果，这是什么原因？
Bevfusion uses the example of testing demo/nucenes with multi_modality_demo.Py. If an image is given as a fake input and the lidar is kept as true, the result is basically correct. However, if the lidar is given as a fake input and the image is true, there will be no output result. What is the reason for this?

Open source status

• The model implementation is available
• The model weights are available.

Provide useful links for the implementation

No response