Why can't the depth buffer in the g-buffer be used?

[YUJIAJING0408]

I tried to find the documentation, but found that only the albedo and normal of the g-buffer can be used as input, is there no depth map because the model was not involved in the training when it was trained? Or is it that depth information is of little significance for denoising.

[atafra]

The current models indeed weren't trained with depth, so that's the main reason why it's not supported yet. Also, depth has a lower significance than the other features so even if it would be supported, the quality improvement would be quite small. Regardless, support for depth will be added in a future version.