Hey,

I am working on nonparametric Bayes methods for latent feature extraction for biological images where I want to find differences in populations. To compare my models, I would like to compare the features with convolutional Autoencoders.

Unfortunately, I get really blurry results. My images show animals and are 147x147x1 pixels - there are minor differences in the image classes. I adapted examples for mnist and other (e.g.: cars[1]), but I still get blurry images, where I can see the animals shape bot no structure on the bodies.

How can I define the architecture for my dataset? Is a convolutional autoencoder the right tool for my application?

[1] https://github.com/foamliu/Conv-Autoencoder