Fine-tuning the prior

[tanner-ducharme]

Does fine-tuning the prior also implicitly train the middle and bottom upsampling layers? Or are those layers left with their pre-trained weights and only the top prior is fine-tuned?

Thanks!