Davide Abati

Hi and thanks for the interest. What kind of additional information do you need? Please note that most implementation details are reported in the supplementary material. Best, D

Hi Cristoph, you can sample from the model by sampling a latent vector from the prior (with ancestral sampling) and then decoding it. A snippet should look like this: ```python...

Great!!! It looked like an over-regularization issue :) Best, D

Hi, what optimizer are you using? Secondly, I see you are printing the running loss (the sum of the loss of the batches so far). Are you sure the loss...

Hi @learnermaxRL I would advise using Adam for optimization. As for the latter error you reported, it is likely to be due to a singleton batch (i.e., a batch with...

Negative values are not undesirable per se, as long as it is a concious choice. What is the range of your input images? D

I would advise standardizing input images. E.g., try making each channel zero-mean and unit std. D

I would advise using a linear activation function for the reconstruction and providing the groundtruth image to the same loss in the same range as the input.

The sigmoid is not mandatory. I would advise not to use it. The rest of the code seems fine. Would an [0-1] input deliver those reconstructions? D

Does the reconstruction loss go down?