search

abhaydoke09 / Bilinear-CNN-TensorFlow

This is an implementation of Bilinear CNN for fine grained visual recognition using TensorFlow.
191 stars 72 forks source link

confuse the tf.einsum() operation? #16

Closed z-huabao closed 5 years ago

z-huabao commented 6 years ago

Hi, I'm confuse with the differents between the code self.phi_I = tf.einsum('ijkm,ijkn->imn',self.conv5_3,self.conv5_3) and self.conv5_3 = tf.transpose(self.conv5_3, perm=[0,3,1,2]) self.conv5_3 = tf.reshape(self.conv5_3,[-1,512,784]) conv5_3_T = tf.transpose(self.conv5_3, perm=[0,2,1]) self.phi_I = tf.matmul(self.conv5_3, conv5_3_T) which in bcnn_DD_woft.py and bcnn_finetuning.py respectively.

They look like the same operation and should be the same operation I think.

abhaydoke09 commented 6 years ago

If you use einsum, then you don't need to flatten/reshape the multidimensional array to carryout the matrix multiplication operations.

https://stackoverflow.com/questions/26089893/understanding-numpys-einsum

https://obilaniu6266h16.wordpress.com/2016/02/04/einstein-summation-in-numpy/

z-huabao commented 6 years ago

@abhaydoke09 Thank you very much for your answer! einsum is really useful!

JUSTDODoDo commented 6 years ago

hello， I still have a problem. After running the second part of the whole model, I will finish training. It seems that the final model is not saved in the code. Why is this done in the absence of the training model? Can you give me some details?