Facebook today open three new AI image segmentation (Image Segmentation) software, namely,DeepMask,SharpMask AndMultiPathNet, Three tool cooperate with each other to complete a full segmentation of image recognition processing, DeepMask generate initial object mask, SharpMask optimize these mask, and finally by MultiPathNet to identify these mask framed objects. SharpMask has followed BSD license on GitHub open source.
Facebook Artificial Intelligence Research Laboratory (FAIR) previously discussed in academic papers in over more than open source image segmentation (Thesis 1,Articles 2,Articles 3). Image segmentation is not only able to identify pictures and videos of the people, places, objects, and even determine their exact location in the image (accurate to pixel level), in order to do this, Facebook uses an artificial intelligence technology & mdash; & mdash; machine learning, which is to train artificial neural networks with large amounts of data, and continuously improve their process determines the accuracy of the new data.
Facebook has been an active promoter of open source technology, artificial intelligence, image segmentation before the three open source software tools, Facebook also worked inTorch Open source some powerful learning tool depth.
Deep learning is the tech giant's competitive technology positions, including Apple, Baidu, Google and Microsoft have invested heavily andCOCO Expand on this image recognition arena of intense competition.
According Facbook presentation, image segmentation technique for improving the social software is of great significance, such as the computer automatically recognizes image objects, which can greatly improve the accuracy and efficiency of the image search, even if they're not adding artificial label. For visually impaired users, the computer can give them even read out the contents of the picture.
Facebook Artificial Intelligence Laboratory scientist Piotr Doll & aacute; r in the blog that: The next challenge image recognition technology is video recognition, in this regard Facebook computational vision technology has made some progress, it is possible to view the video at the same time to understand and differentiate the video objects, such as cats or food. Real-time video capabilities to distinguish objects will greatly improve the recommendation accuracy Facebook live video content, and with the technology level, future machines will be able to give real-time description of the scene based on temporal and spatial variation, objects and actions.