Main image of article Facebook Open-Sources A.I. Computer Vision Tool
If you’re curious about computer vision (and if you work with anything related to images, cameras, or mobile, you should be), Facebook’s artificial intelligence research division has open-sourced a new toy for your programming pleasure. Detectron, a project begun by Facebook AI Research (FAIR) in July 2016, is a codebase that facilitates computer-vision projects. Originally intended as an object detection system built on Caffe2, it subsequently evolved into this vital support element, according to a posting on Facebook’s official research blog: “The codebase has matured and supported a large number of our projects, including Mask R-CNN and Focal Loss for Dense Object Detection, which won the Marr Prize and Best Student Paper awards, respectively, at ICCV 2017.” In addition to computer-vision research, Detectron is useful for training custom models for augmented reality (AR), community monitoring, and other initiatives. “Once trained, these models can be deployed in the cloud and on mobile devices, powered by the highly efficient Caffe2 runtime,” the Facebook blog added. “Our goal in open sourcing Detectron is to make our research as open as possible and to accelerate research in labs across the world.” (Although the social network has aggressively pushed virtual reality via the Oculus platform, it has also taken big strides in AR, including the launch of an AR Studio in late 2017.) Tools such as Detectron are great for hardcore A.I. researchers who have years of training and experience. The big question facing the A.I. industry as a whole is whether tech firms will begin “democratizing” their A.I. tools for people without that background. Google recently took a step in that direction with the unveiling of Cloud AutoML, a platform designed to automate the generation of customized machine-learning models. In theory, companies and developers will have the ability to create these models via simple drag-and-drop. “While Google has offered pre-trained machine learning models via APIs that perform specific tasks, there’s still a long road ahead if we want to bring A.I. to everyone,” Fei-Fei Li and Jia Li, chief scientist and head of R&D, respectively, of Google’s Cloud AI division, wrote in a lengthy Jan. 17 blog post announcing Cloud AutoML. Will other companies involved in A.I. research—including Facebook, Amazon, and Microsoft—allow Google to move unopposed into the “beginner A.I.” space? When you consider the importance of a healthy A.I. ecosystem to these firms’ near- and long-term strategies, the answer is “probably not.” They will continue to produce tools such as Detectron, but you may see more platforms along the lines of Cloud AutoML emerge, as well. That’s good news for tech pros who understand a little bit about artificial intelligence, and want to build apps that leverage it, but don’t have the time to become experts.