Google Research’s RAWNeRF, a constituent of the multiNeRF research, promises results that can put it ahead of any other noise-reduction tool. The RawNeRF tool uses artificial intelligence to read images and add higher levels of detail to photos taken in darker and low-light conditions.
According to a Cornell University paper, NeRF produces a scene representation so accurate when optimized over many noisy raw inputs that its rendered novel views perform better than dedicated single and multi-image deep raw denoisers that run on the same wide baseline input images.
NeRF is a neural network tool that is capable of reconstructing accurate 3D renders from a group of images. As per Ben Mildenhall, a researcher at Google, the NeRF is built to work best with well-lit scenarios.
However, when tried with images that are taken in low-light conditions, the results compromise on details and are noisier. The issue can be solved with denoising tools, further losing details.
Meanwhile, the algorithms are run on RAW images in the RAWNeRF, and AI is tasked to decrease the noise captured by the sensors while maintaining the detail, letting us see in the dark.
Google said that the RAWNeRF is more capable of reducing noise than any other technology. It can change the camera position to show the scene from different angles, or change tone map, exposure, and focus with accurate bokeh effects.