Is it possible to replace image preprocess(normalization) with a 1x1 conv layer?

[RunshengZhu]

Thanks for this wonderful work~ @chuanqi305 Recently I was using this repo translating tensorflow mobilenetv2-ssd model into caffemodel, everything was ok. But i noticed that there was a preprocession before inference of caffe model in the demo which would cause some consume on cpu. I wonder if it's possible to replace it with a 1x1 conv layer. Waiting for response, thanks!

[xinxin12345]

Thanks for this wonderful work~ @chuanqi305 Recently I was using this repo translating tensorflow mobilenetv2-ssd model into caffemodel, everything was ok. But i noticed that there was a preprocession before inference of caffe model in the demo which would cause some consume on cpu. I wonder if it's possible to replace it with a 1x1 conv layer. Waiting for response, thanks!

Have you implemented the replacement 1×1 conv layer,and what is the effect?