wangshy31 / densely-connected-cnn-with-multiscale-feature-attention Goto Github PK

Is there a validation set in the experiment? Or how do you apply 10 fold cross validation on MR?

wangshy:
hello! After reading your paper, I want to ask about how to keep the multi-scale features from different layers having the same size. Your code implemented by Caffe puzzled me a lot. So, can
you just explain me in a simpler way how your ideas work? Just like giving me a example. For example, Suppose, the output of first land the second layer is 128 * 56. In the third layer, it receives 128 * 112. So, how to keep its output as 128 * 56?
Thank you!

I have try this, but fail to build the caffee by "make all".
It may the cause of version of the proto, is there any tips for this issue?

Caffe is not very well, do you have the realization of the tf or pytorch？

Shy:
Very interesting framework! In your paper, the last paragraph in subsection 4.4, you mentioned that Table 3 lists a few phrases ranked by the product of normalized neural responses and etc.. I don't understand the meaning of "normalized neural responses" here. Can you explain it? Thx !