When I try to reimplement the DenseASPP, I refer to both the official code and this repo, but when it comes to DenseASPP head, I notice the order of conv and norm in DenseASPP block is different.
In official code, it's norm->relu->conv->norm->relu->conv, in this repo, it's conv->norm->relu->conv->norm->relu.
Just the preact form of conv and postact form of convolution, is that mean the postact form performs better in semantic segmentation after your experiment?
When I try to reimplement the DenseASPP, I refer to both the official code and this repo, but when it comes to DenseASPP head, I notice the order of conv and norm in DenseASPP block is different.
In official code, it's norm->relu->conv->norm->relu->conv, in this repo, it's conv->norm->relu->conv->norm->relu.
Just the preact form of conv and postact form of convolution, is that mean the postact form performs better in semantic segmentation after your experiment?