30th April 2018 New AI algorithm reconstructs images Graphics chip firm Nvidia has developed a new deep learning method to automatically reconstruct images with missing pixels. The AI uses a process called "image inpainting", which could be implemented in photo editing software to remove unwanted content, while filling it with a realistic computer-generated alternative. "Our model can robustly handle holes of any shape, size location, or distance from the image borders. Previous deep learning approaches have focused on rectangular regions located around the center of the image, and often rely on expensive post-processing," the researchers state in their paper. "Our model gracefully handles holes of increasing size." To train their neural network, the team first generated 55,000 masks of random streaks and holes of arbitrary shapes and sizes. They also generated 25,000 for testing. These were further categorised into six groups based on sizes relative to the input image, in order to improve reconstruction accuracy. Nvidia will be presenting its latest research at the International Conference on Learning Representations (ICLR), being held from 30 April – 3 May 2018 at the Vancouver Convention Center, Canada.
Comments »
If you enjoyed this article, please consider sharing it:
|
