Why you use the labels to generate the 3D centres?

[deepmeng]

I see the function " evaluate3d_detection()" in file “evaluation.py ”. And I find that you obtain the 3D centres based on the 2d labels. Also, the objects' 3D sizes are related to the samples' labels.

[deepmeng]

In file "predict.py", some preprocessing operatings which are before running the model for 3d prediction also refer to 2d labels.