Google camera against the sky AI night scene photo new technology RawNeRF unveiled: perfect noise reduction, but also synthesize 3D view angle

Recently, the domestic online video of RawNeRF, a new Google AI night photography technology, can be perfect noise reduction, accurate restoration of night photos, and can synthesize a number of 2D photos 3D vision, and can adjust the focus.

The paper on the technology was published in November 2021 and also has a home page on Github.io, titled "NeRF in the Dark: High Dynamic Range View Synthesis from Noisy Raw Images".

As the name suggests, the technology can be processed by NeRF (Neural Radiation Field) high noise images taken at night, so that the Raw images full of noise synthesis of high dynamic range view.

NeRF is a simple fully connected neural network that uses information from 2D images as training data to restore 3D scenes with volume.

While the original NeRF uses tone-mapped, low dynamic range LDR images as input, Google has modified NeRF to train directly on linear raw images, preserving the full dynamic range of the scene, effectively turning RawNeRF into a multi-image noise reducer capable of combining information from tens or hundreds of input images.

In addition to changing the camera perspective, RawNeRF can adjust the focus, exposure and tone mapping in post, that is, as demonstrated in the video.

Partners should be amazed by the powerful "night vision" function of Google Camera, if RawNeRF can be used in Google Camera, then perhaps cell phones can also shoot clear 3D visual images at night, of course, not yet.