Back in the day it was pointed out that in TV shows and movies, being able to pull up a grainy, low-res photo and blowing it up to make it sharper was fiction. However fast forward to today, that fiction is slowly becoming a reality, thanks to the development of AI that can help to enhance low-res photos.
We’ve seen various efforts, such as Let’s Enhance and also from Google who taught an AI how to edit photos, which is why it isn’t surprising to see that there are other researchers who are working on similar tech. Dubbed Deep Image Prior, this is another neural network that can not only enhance your low-res photos, but can also apparently fixed corrupted image files as well.
What’s interesting about this particular method is that unlike other methods that require the AI to be trained/taught first, Deep Image Prior instead uses the image itself as a guide. According to the research paper, “Our method uses a randomly-initialized ConvNet to upsample an image, using its structure as an image prior; similar to bicubic upsampling, this method does not require learning, but produces much cleaner results with sharper edges.”
It is an interesting initiative and no doubt more efficient than if a human were to do it themselves, but when it will be implemented by developers into mainstream photo editing software remains to be seen.