Closed dsy412610 closed 8 years ago
I do a simple experiment, creating a model with 7 stage of pooling (2x2). Two different kind of pooling don't have any computation time difference. Can you provide codes to reproduce this phenomenon? P.S. I use TensorFlow backend.
@joelthchao @moothes Thank you very much! Average pool is only available in Theano when cudnn is there, but I didn't use cudnn.
Using AveragePooling may increase computation cost. But after I replaced MaxPooling with AveragePooling, training time became 40 times longer than before. Is there anything wrong? Thanks!