I dont get the input and output shapes of pfe.onnx and rpn.onnx

[mihsamusev]

Hi, i have a stupid question.
Can anyone explain the input shapes to the pfe.onnx an rpn.onnx. I have difficulty understanding how to apply it for my point clouds. So what i see is that i somehow need to create pillars and shape x,y,z to individual [1, 1, 12000, 100], and i dont see how it corresponds to the decsribed tensor (D, P,N) with 9 features D, pillars P and points per pillar N. Furthermore, what is the output of pfe.onnx and and most importantly the input and output of rpn.onnx. Just not managing to relate the given neural network shapes to the expected bounding boxes.

thanks