As the worlds of artificial intelligence and digital photography collide, we begin to see some mind-blowing technology.
The current research in the rotation with low-resolution photos high-definition photos, you can let your jaw — it starts to cross over into the realm of sci-fi.
What you see above is not a mockup of what this enhancement technologies can do in a day: it is a current example of an AI-powered extension is performed by a new algorithm called EnhanceNet-PAT developed by scientists at the Max-Planck-Institute for Intelligent systems in Tübingen.
The example started with this original photo of a bird:
The scientist took the photo and I lost track of this low-resolution version, in all the fine details are:
What are the fine details are of course different in the enhanced version compared to the original, but what is lost in accuracy for realism.
“While the pixel-perfection, the sacrificed, the reward is a better result,” the Max-Planck-Institute writes.
Here are some more examples of low-res photos (on the left) and the improved high are manufactured-res versions by the algorithm (on the right):
Great photos of the little one is a field, known as single-image super-resolution (SISR) technology. Yesterday, we featured a free website with the name “Let’s Enhance’ s, and use SISR-to strategy for this type of image processing as a service.
Traditional upscaling technologies have tried to include the missing Pixel and details, looks through the calculation, based on the surrounding pixels. The results of these types of strategies are blurred and unsatisfactory. What the scientists are researching now is with the machine learning to “learn” what features should have low-resolution photos, also through the study of the original high-resolution versions.
Once formed in this way, the algorithms are able to, new low-res photos, and make a guess as to what would look like an “original” high-res version of the photo.
“From the location of the recognition and generation of patterns in an image with low resolution and the application of this pattern in the upsampling is a thinking process, EnhanceNet-PAT, like the birds feathers should look, and adds additional pixels in the image with lower resolution,” of the Max-Planck-Institute says. “You could say that the technique has created its own reality. For most viewers, the result is very similar to the original photo. The image of the bird is good to decorate the photo album.”