CODEWITHSUNDEEP
pythonkerasdeep-learningconv-neural-networkautoencoder
Bakhti
Bakhti

Reputation: 11

Using non-square input matrix for convolutional autoencoder

Is it possible to train convolutional autoencoder (CAE) with non-square (rectangular) input matrix? All the tutorials and resources I have studied on CAE seems to use squared images. The data I am working with is not image. I have hundreds of single cells and for each cell there is a matrix (genomic data) with thousands of genes in rows and hundreds of bins in columns (genomic region of interest for each gene divided into the bins of equal size).

I have tried some models with Keras, but size of input in the encoder part of the model is always different than the size of output matrix in the decoder. So it is giving error. Can someone help me how to solve this problem?

Upvotes: 1

Views: 600

Answers (1)

mousa alsulaimi
mousa alsulaimi

Reputation: 336

it is hard to tell what is the issue here since no sample code is provided . However most propably your Matrix is diminsions are odd ( example 9×9 ) or becomes odd while pooling. tofix this issue you need to either pad your input to make the matrix diminsions even . our crop the decoder lsyers of your autoencoder to have a matching output size

Upvotes: 0

Related Questions