question about Vb loss

[ljw919]

I want to perform an "Upsampling" task, so I choose the low-resolution image as the condition. While the problem is that, if I set learn_sigma = True, the _vb_terms_bpd will be calculated with model_kwargs set as None (default) in the code, does it means when calculates vb loss, I dont need to add the condition, just perform an image generation forward without condition? Sincerely looking for your answer.

[ljw919]

However, the code include model_kwargs in _vlb_terms_bpd while choose the KL loss, so i guess whether the model_kwargs is forgetten in the MSE loss (learn_sigma).