Facebook wants you to build machines that can see like humans, gives away its AI
Facebook has open-sourced its computer vision tools for everybody.
Facebook announced on Thursday that it is going to give away its image processing Artificial Intelligence. A set of three software namely DeepMask, SharpMask, and MultiPathNet which were developed by the Artificial Intelligence Research of Facebook have been open-sourced on GitHub for the general public.
All of these software can be used by computers to contextualize what exactly is in a picture. The software will help Computers translate an image much likely as a human would do. Essentially, it is not possible for a computer to understand an image. For a computer, an image is merely a set of different colors and transitions. At most, it can categorize the colors of an image and can assign a number to each color. That is where Machine Learning comes in. Computer is being fed with large amounts of data (Neural Networks). The data contains numerous examples which teach the computer what it is in the picture.
DeepMask and SharpMask are used for segmentation of an image.
- DeepMask can be used by a computer to ‘segment’ an image. Segmentation is the process which separates the contents of an image. It outlines the contents of an image.
- When the contents of an image are being segmented, SharpMask is used to help computer refine its selection. The percentage of error is cut by a huge margin as SharpMask refines a segment.
- MultiPathNet is actually the software which categorizes different segments of an image. A series of tests are performed which help a computer to classify the segments of an image with the help of available neural networks.
The AI will help researchers to build smarter machines and accelerate the research process.