AMD presents new method of handing indirect lighting off to an image generation model: 'frame-by-frame' solution could be part of AMD's answer to DLSS 5 someday
"We try to solve global illumination as an image generation problem," says one of the researchers behind the method.
AMD has unveiled a novel method for incorporating indirect lighting into image generation models, potentially paving the way for a future version of AMD's FSR Diamond technology. The researchers utilized a one-step diffusion model to produce realistic lighting effects on a frame-by-frame basis, eliminating the need for computationally intensive techniques like path tracing.
By training the model on a diverse dataset of 31,000 frames rendered using Blender Cycles, it learned to generate lighting that aligned with the scene's geometry and materials. To maintain consistency and avoid flickering, the researchers employed a temporal VAE, which stores a history of previous outputs to ensure lighting remains coherent across frames.
The final result is compositing with the original frame, which softens any errors in the generated lighting. AMD claims the method can run in real-time with a frame generation time of 0.29 seconds, producing a resolution of 512 x 512 on an RTX 3090 GPU. Currently, this technique only addresses the global illumination step, leaving other aspects of rendering untouched.
Nonetheless, it represents a significant step forward in the field of neural rendering and could potentially disrupt the competitive landscape in GPU-accelerated graphics.
Written by urgent.news from PC Gamer's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
