Yi+ Tops the Chart for "Labeled Faces in the Wild" (LFW): Surpassing Tech Giants in Face Recognition Performance

LFW was established by the University of Massachusetts in 2007 to evaluate the performance of face recognition algorithms under unconstrained conditions. It is also one of the most widely-used evaluation sets in this field. Thus far, dozens of global teams have submitted more than 80 sets of test results. They include Google, Facebook, Microsoft Research Asia, Baidu, Tencent, SenseTime, Face++, Hong Kong Chinese University and are not limited to, other corporate and academic teams. Many cutting-edge methods from earlier times played a crucial role in the development of the face recognition technology. They also served as the litmus test for evaluating its performance.

In this LFW participation, Yi+ adopted an unrestricted, labeled external data protocol. The system consists of face detection, face alignment and face descriptor extraction. Yi+ used a multiple loss functions and training datasets to train the CNN model, which contained about 10M images from multiple sources, including 150,000 individuals (the training dataset had no intersection with the LFW). In the testing, Yi+ used the original LFW image and applied a simple L2norm. The similarity between pairs of images was measured with the Euclidean distance and eventually achieved excellent results.

Official Website of Labeled Faces in the Wild (LFW)

Face recognition is one of the core products of Yi+AI. In recent years, the company has made continuous progress in new technologies such as artificial intelligence and big data. Now, it could complete face detection, key point detection, and face attribute detection - accurately and swiftly, and realize the unique attributes of on-screen characters, such as gender, age, race, emotion, face value, sexiness and fashion. This thus supports the large-scale recognition of celebrities at home and abroad. Using deep learning, large-scale face search and comparison for face clustering purposes and sensitive target monitoring can also be accomplished.

Face Comparison Technology

At present, Yi+AI has integrated face recognition technology into three major solutions: "Television", "Camera" and "Marketing". This has further branched out to virtual makeup, smart GIF advertising, content recommendation and diversion, broadcast safety and control, face recognition advertising, membership system and other applications.

Yi+ (Beijing Moshanghua Technology) is a leading computer vision engine service provider, providing enterprises with visual content intelligence and commercialization solutions. The company started with the recognition of commodity objects. After four years of development and technology accumulation, the current detection, recognition, segmentation, tracking, search and recommendation of scenes, general objects, products and faces in images and videos have reached sophisticated levels in the world.

As Yi+ recognizes that the development of science and technology is inseparable from the deep cultivation of professional fields, it continues to seek talents and professional scholars from well-known universities both at home and abroad. Currently, the team members come from many top companies and universities, such as Stanford University, Imperial College of London, Yale, Princeton, National University of Singapore, Nanyang Technological University, Tsinghua University and other famous enterprises like Google, Microsoft, IBM, Intel, Alibaba, Tencent, Baidu and Huawei.

Going forward, Yi+ will combine the advantages of existing technologies in the spheres of public safety, information security, and intelligent marketing. The deep accumulation of our teams in deep learning and data analysis will give strength to face recognition for the creation of a smart city. This is in line with the vision to provide people with artificial intelligence services using science and technology.

SOURCE Yi+ (Beijing Moshanghua Tech Co.,Ltd.)