zeakey opened a new issue #12838: Module.predict(): different behaviour with caffe's net.forward()? URL: https://github.com/apache/incubator-mxnet/issues/12838 It seems that mxnet's `Module.predict(data)` automatically performs a channel switch. I have two identical networks, say `mx_net` and `caffe_net`, in mxnet and caffe respectively, I've mannually checked all the parameters are identical. I found that he outputs of the two nets are different by: ``` mx_output = mx_net.predict(data) caffe_output = caffe_net.forward(data) ``` Interestingly, if I manually switch channel of the data, I got identical outputs: ``` caffe_output = caffe_net.forward(data) data = channel_switch(data) mx_output = mx_net.predict(data) ``` So is it real that MxNet performs an automatic channel switch inside the `Module.precit()` routine? If so: firstly, this is an elegant design and really friendly to beginners, and second, can some one point out where does this happen in the source code?
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
