keras2caffe icon indicating copy to clipboard operation
keras2caffe copied to clipboard

Published 20 hours ago • verified publisher icon uhfband

convolution layers of keras and caffe has different output

Open xiaofeiwu opened this issue 6 years ago • 5 comments

It seems that the output of convolution layers of keras and caffe has different output even when the input the and weights are all the same. It may be caused by the padding implementation in tensorflow and caffe. Does anyone meet the same issue?

xiaofeiwu avatar Jun 06 '18 07:06 xiaofeiwu

Yes，I met the same problem.Did u solve the issue?

nerddd avatar Sep 14 '18 01:09 nerddd

I think pooling layer has the difference between caffe and tf

nerddd avatar Sep 14 '18 07:09 nerddd

It seems that the output of convolution layers of keras and caffe has different output even when the input the and weights are all the same. It may be caused by the padding implementation in tensorflow and caffe. Does anyone meet the same issue?

I met the same question.Did you solve the problem?

to-be-snail avatar Feb 21 '19 07:02 to-be-snail

I think pooling layer has the difference between caffe and tf

I met the same question.Did you solve the problem?

to-be-snail avatar Feb 21 '19 07:02 to-be-snail

So, bascially, the transfered caffemodel can not get the same results as the keras .h5 model predict?

DanielXu123 avatar May 14 '20 03:05 DanielXu123