chasingw

confused about Dilated convolution replace maxpooling

1

hello, I'm confused about this change in class PillarFeatureNet https://github.com/SmallMunich/nutonomy_pointpillars/blob/68d3181d35702608d410e948608e9815815677b9/second/pytorch/models/pointpillars.py#L52-L73 use conv replace maxpooling op although the shape are same, can this change equivalent? do you know some reference material...

Performance on livox_dataset

2

how about eval result on the livox dataset