To restore old or damaged photos often spent a lot of time and effort, and every lost fragment is restored manually almost pixel by pixel. But it is not the first year are developing an automated system designed for this purpose. And recently, a team of engineers NVIDIA introduced a neural network, able to finish the lost item and remove the “extra” like a watermark. Moreover, the technology of NVIDIA at the moment has no analogues in the world.
The algorithm was developed and submitted by the engineer of the Guilin Liu. The neural network now NVIDIA has a competitor in the face of Adobe, which is working on similar technology. Technology, however, operates differently. The creators of Photoshop use AI, who in the course of image processing borrows a context-sensitive display of the neighboring pixels, and then completes the image in accordance with this. This approach is not always convenient, because the missing fragment can contain multiple colors. For example, the edges of it will be red, and in the center in blue. The system sees only the edge of the object and on the basis that “fills” all in red.
The algorithm from NVIDIA works differently. It analyzes the context of a particular area, and the big picture, identifying patterns and relationships between the objects. Recognizing the object, the neural network “assumes” that could be in the missing places, and then “draws” all the way, to maximize results and fit not only in colors, but in the context of the original picture. For example, with blue and red object, the algorithm will “look” at the whole picture and if you see similar fragments, will try “to finish” the missing region so that it is as similar as possible to the original.
When the technology is ready, it allows to recover photos for a couple of minutes. But now, as you can see in the video is available below, AI copes, “painting something” eyes in the portrait of a man.