[ad_1]
If you have already taken a picture in low light, you have probably come across the granular effect that can dilute the finished product. A new AI tool could be an incredibly easy way to remove this noise, with the ability to automatically produce a clean image after analyzing only the corrupted version.
The IA was built by researchers from Nvidia, MIT and the Finnish Aalto University, with the aim of improving recent work in the field of machine learning and image processing. This has led to neural networks capable of scanning a noisy image with a clean version, and then using automatic learning to bridge the gap.
Their solution, dubbed Noise2Noise, is able to produce an image without noise without ever seeing a clean original version, automatically removing artifacts, noise and grain. The team developed the tool by training it on a catalog of 50,000 images, and tested it on three different data sets including images of different types.
"It is possible to learn to reproduce signals without ever observing clean signals, sometimes exceeding performance with the help of clean copies," the researchers note in their article. . "The neural network is up to the cutting edge methods that use own examples – using precisely the same training methodology, and often without appreciable inconvenience in the time of training or performance."
could one day make Instagram a lot more beautiful, in the field of science, it could also have a significant impact. Medical images taken by MRI and spatial images taken with scientific instruments are two typically granular examples that could benefit from this type of technology.
"There are several real-life situations where getting clean workout data is difficult: low-light photography (eg astronomical imaging), physical rendering, and magnetic resonance imaging" says the team. "Our proof-of-concept demonstrations pave the way for significant potential benefits in these applications by removing the need for a potentially painstaking collection of clean data. There is no free meal – we can not learn to acquire features that do not exist "
A research paper describing the technology is available online, and the work will be presented at the International Conference on machine learning in Stockholm, Sweden, this week.The video below provides more examples in action.
View Gallery – 4 Images
[ad_2]
Source link
