Detection head (SSD) looks like a single convolution

[ahmedius2]

Hi,

I looked into what layers are there inside the rpn.onnx network. I can see the backbone part mentioned in the paper. However, it looks like the detection head (SSD) part consists of only three parallel convolutions:

I would expect a a much longer network as SSD is longer than 1 layer. Did authors of Autoware version done that intentionally?

Best