TRENDING ON ONEINDIA
- Four-Storey Building Collapses In Gurugram; Many Feared Trapped
- Virat Kohli To Be Rested For Last Two ODIs And T20I Series Against New Zealand
- New Mahindra Thar Model’s Spy Pics Out Again — Is It The Largest Indian Off-Roader Ever Made?
- PayTM Mall On Republic Day — Samsung Carnival Offers
- Best Low-Priced Stocks To Buy For 2019
- Ajay Devgn Shares His Opinion On #MeToo Movement
- All You Want To Know About Toothukudi
- Deepika Padukone Adds Another All-black Ensemble With This Airport Attire
When it comes to graphics, it's Nvidia who steals the stage away most of the time. This time also a team of researchers have developed something interesting for the users of all age group. According to some sources from the web, a team of researchers which is led by Guilin Liu have recently developed a new technique that makes use of AI (artificial intelligence) to restore or repair damaged and corrupt images that are hard to restore otherwise.
Apart from restoring pictures, this will also help in editing the image by removing contents and filling in the holes. The method of restoring the images involves a process named 'image inpainting'. This process uses realistic computer-generated alternative in order to replace the unwanted parts of an image.
The research team has submitted a research paper related to the technique that quotes:
"Our model can robustly handle holes of any shape, size, location, or distance from the image borders. A previous deep learning approach has focused on rectangular regions located in the centre of the image, and often relies on expensive post-processing...Further, our model gracefully handles holes of increasing size."
The report further suggests that the researchers were able to train the neural network by generatingb5,116 masks of random holes of arbitrary shapes and sizes. The team of the researchers then also generated 25,000 similar holes for testing. The masks of the random steaks and holes were further divided into six categories that are based on their size relative to the input image. Following this, the team made use of Nvidia Tesla V1.00 GPUs and the cuDNN-accelerated PyTorch deep learning framework. The aim behind this was to train the neural network by applying masks which are generated by the ImageNet, Places2 and Celeb- HQ datasets.
It is interesting that Nvidia could come out with such a technology that can help users relive their memories which otherwise would have been corrupted or damaged. We are embedding the inpainting video and you can see how the results displayed are pretty good. That said you can check out the video below and let us know what you think about it in the comments section.