pushkar-khetrapal / fast-cnn Goto Github PK

Hi! I am trying to create a FFT convolution layer in a keras/tensorflow NN. I have tried to replicate your code, but there is an issue in fft_forward due to the weights reshape into a (3,3) array.

how find and print Accuracy for fft_model and compare it with CNN by using model.fit .

where when run the following error appears

AttributeError Traceback (most recent call last)
in ()
2
3
----> 4 output.compile(loss='binary_crossentropy',
5 optimizer=RMSprop(lr=0.0001),
6 metrics=['acc'])

AttributeError: 'tuple' object has no attribute 'compile

thank you

The claim in this article and the notebook is quite misleading.

pyTorch CNN implementation already uses the cuDNN or a faster fbfft module. For smaller kernel sizes, the Winograd algorithm is even faster.

Your comparison is against convolution done in the position space, using a python for loop.