One of the most important tasks that one has to take care of while working with image editors is object selection. Users have come up with tricks to accelerate the process but there was no good automated feature that could do the same with human-level accuracy. However, MIT’s Computer Science and Artificial Intelligence Lab (CSAIL) put an AI-assisted image editing tool to the task and the end results look quite promising. Researchers have named the new tool Semantic Soft Segmentation (SSS), which is based on soft segments that automatically fuse high-level information from a neural network with low-level image features. The research thesis states that the new feature makes some image editing tasks easier, which would otherwise consume a skilled artist’s time and labour.
Detailing the method, the MIT research paper reads, “We have shown that by carefully defining affinities between different regions in the image, the soft segments with the semantic boundaries can be revealed by spectral analysis of the constructed Laplacian matrix. The proposed relaxed sparsification method for the soft segments can generate accurate soft transitions while also providing a sparse set of layers.” This method uses neural networks to determine image features and detect soft edges in an image but soft transitions can cause two dissimilar objects or the background and an object to share pixels around edges.
The AI factors this in into account and handles the repetitive work of separating the layers autonomously. Alongside image processing, AI is being applied to many other areas as well. Elon Musk’s OpenAI managed to beat Dota 2 players in a one on one match last year and it recently went up against professional players at ‘The International’, which is the gaming title’s annual championship tournament. The AI-based bots lost their first fight against human players but there are still two more bouts remaining, which will take place over a span of two days.
Hello, my name is Jack Sparrow. I'm a 50 year old self-employed Pirate from the Caribbean.