how Google’s new AI will kill the game

Although Google has made great efforts to improve the photo sensors of its smartphones on the Pixel 6 and the Pixel 6 Pro, the strong point of the Mountain View company remains the software. Indeed, it invests heavily in artificial intelligence and thanks to its advances in this area, it is able to offer important features to improve the quality of photos and videos taken by its smartphones.

And to this is added, on the latest Pixels (including the Pixel 6a), the use of a chip designed by Google which is optimized to meet the needs of the company’s software in terms of AI.

In any case, Google is not resting on its laurels, since its researchers continue to experiment with new techniques to improve the quality of photos or videos. And precisely, recently, the Google Research team presented a new technology that later could revolutionize night photos.

Impressive results for noise cancellation

Called RAWNeRF, it is inspired by an already known technique, called NeRF. According to the explanations of our colleagues from 9to5Mac, a NeRF or neural radiance field is a neural network (therefore, AI) which takes 2D images to generate 3D scenes.

What distinguishes Google’s RAWNeRF technology is that it uses raw or RAW images. And apparently, in addition to generating very good quality 3D scenes, the technique also has the potential to improve images taken in the dark.

Indeed, on night photos, RAWNeRF is able to offer an impressive noise suppression function.

“When optimized on many noisy raw inputs (25-200), NeRF produces such an accurate scene representation that its newly rendered views outperform single- and multi-frame dedicated deep raw denoisers run on the same input frames. wide bases. As a result, our method, which we call RawNeRF, can reconstruct scenes from extremely noisy images captured in near darkness.”, can we read in the presentation of this project. In other words, RAWNeRF would be more efficient than some software dedicated to denoising.

“Training directly on the raw data effectively turns RawNeRF into a multi-image denoiser capable of combining information from dozens or hundreds of input images”can also be read further on.

See for yourself

Below you have a video presentation of the technology. From 2:33, the video shows examples of noisy images taken in the dark, then the result obtained after the action of RAWNeRF. In addition to having eliminated noise, the technology provides very detailed images.

And that’s not all, since it is still possible to perform processing, such as creating a bokeh effect, or to play with other parameters on the scene.

Unfortunately, at the moment, this is only research. And as 9to5Google explains, it is not certain that RAWNeRF will land on Pixel smartphones, since the software uses several raw or RAW images which are large.


Leave a Comment