Is something wrong in inner_product_layer.cpp? #5768

Closed
gauss-clb opened this Issue Jul 12, 2017 · 1 comment

Comments

Projects
None yet
1 participant

gauss-clb commented Jul 12, 2017 edited

https://github.com/BVLC/caffe/blob/master/src/caffe/layers/inner_product_layer.cpp#L33
I think the shape of weight should be K_* N_ if transpose_=false, because the shape of bottom blobs is reshaped to M_* K_, so is it wrong from line 33 to 39?

But weight are initialized randomly so the effect of K_* N_ is the same as N_* K_, because we use 1-dimension array to store weights.
But I think the code should be correct at the shape of blobs logically. Do you agree with me?

Oh, I'm sorry.
https://github.com/BVLC/caffe/blob/master/src/caffe/layers/inner_product_layer.cpp#L89
transpose_ ? CblasNoTrans : CblasTrans, the transpose_=true means CblasNoTrans rather than CblasTrans,
it is amazing.

gauss-clb closed this Jul 16, 2017

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment