jonathanalis said:
Well, I would guess that 4K footage of a game is needed to the training of the DLSS network for reconstructing 4k frames from that game. Well, not exactly needed, but would improve the image quality for specific content. I mean, if you train the network with a specific art-style footage, the reconstruction will tend to recreate this style. You could have a network trained in general images, but sometimes it would recreate non intended content. A cell-shaded game could lose its cell-shaddy details if reconstructed by a network trained with realistic content, for instance. |
16K source image.
@14:37