variationalloss.py

def vae_loss(input_img, output):
    # compute the average MSE error, then scale it up i.e. simply sum on all axes
    reconstruction_loss = K.sum(K.square(output-input_img))
    # compute the KL loss
    kl_loss = -0.5 * K.sum(1 + log_stddev - K.square(mean) - K.square(K.exp(log_stddev)), axis=-1)
    # return the average loss over all images in batch
    total_loss = K.mean(reconstruction_loss + kl_loss)
    return total_loss

Copy-pasting over from the old one, it appears the previous gist + links broke