shoul i add your ConvolutionDepthwise layer to the caffe.proto？

[123chengbo]

I want to use mobilenet for Object detection，so i tried your caffe-implementation firstly。 i have add your files to the caffe （include、src）， i dont konw whether should i modify the caffe.proto， for the new ConvolutionDepthwise ，could you help me？

[ouceduxzk]

same problem here, without doing anything else, the my old version of caffe compiles, but crashes when running , this layer is not well registered

[farmingyard]

@123chengbo
You don't need to modify caffe.proto, because dw conv uses conv's parameters.

[farmingyard]

@ouceduxzk
I think you should use newer version caffe.

[ouceduxzk]

@farmingyard thanks for your suggestion, I have a unrelated question, the newer caffe remove the data reader layer and I can not reuse my customized data_transformer layer, do you have some suggestion about that ?

[farmingyard]

@ouceduxzk
caffe's data related layers are truly complex, i am also confused by them sometimes, i suggest you choose a newer version than yours but not the newest caffe.

[ouceduxzk]

@farmingyard thanks, I tried the newest version of caffe that is before the deletion of the data reader layer, but i still have the same problem

[epleone]

@ouceduxzk Maybe you can use chmod -R 777 .