Want to fix a childhood’s photograph or need to remove someone from your prom photographs? NVIDIA’s new artificial intelligence (AI) can you help you out. Every time, when it comes to restore any incomplete or decade-old photo, the solution is, “Photoshop it!” However, the existing image editing tools are not that intelligent in fixing incomplete photos as they try to borrow content from adjacent pixels. However, NVIDIA has developed an AI for image correction that can determine what should have been there instead of missing pixels. NVIDIA is the company that designs graphics processing units (GPUs) for the gaming and professional markets. According to its official blog, it has developed a deep learning tool that determines the missing portions of the image and not just copy from adjacent pixels. For example, if a picture is missing an eye, the program inserts an eye if the area is largely concealed.
There have been several attempts to develop a perfect image repairing AI. However, they were limited to rectangular sections and focused on the images that are not properly scaled. NVIDIA incorporated “partial convolution” approach, which ensures that output does not limit input values and can work with any size, shape, or location. This enables the tool to create uncanny results, despite the fact that AI does not quite understand the missing part. NVIDIA trained its AI by feeding thousands of hole variations, and AI learned how to repair the photos by testing different sets of holes to restore the photos. However, sometimes the outputs are not flawless, and one may see facial features borrowed from someone else as the program struggles due to insufficient information. However, the AI is incredibly useful to repair a decade-old photograph just by minor touch-ups and saves hours of work. The company believes this AI could end the tedious work of recreating a picture from scratch.
Few months ago, Adobe, American multinational computer software company, developed Deep Fill tool that functions on similar lines of NVIDIA’s AI. Deep Fill uses AI to understand the content in a picture so that the resulting fill would look more realistic and most importantly, sensible. According to Adobe, “Deep Fill is an advanced deep neural network-based image in-painting system that learns to produce images patches more realistic and semantically reasonable and permits interactive customization of result based on users’ inputs.” Similar to Adobe’s Deep Fill, NVIDA’s deep learning AI reconstructs the image, removes people without messing up the background, and completes incomplete facial expressions. However, NVIDIA’s tool is much more sophisticated than Adobe’s Deep Fill.
According to the whitepaper published by NVIDIA, the goal of the research is to develop a model for image in-painting that operates mainly on irregular patterns and generates meaningful predictions. However, the company has not disclosed the date when people will get their hands on the tool. But, for now, it has given a hope for a better and easy future for image editing.