iQIYI releases dataset that enhances facial recognition tech
01 November 2019 13:23 GMT

iQIYI, an online entertainment service in China, today announced that its paper on facial recognition technologies - Unknown Identity Rejection Loss: Utilizing Unlabeled Data for Face Recognition has been accepted by the International Conference on Computer Vision (ICCV), one of the world's top three conferences on computer vision whose accepted papers represent the most advanced research in the field.

Along with the development of deep-learning technology, facial recognition has been a subject of widespread research. Practical applications of facial recognition, however, are still lacking due to challenges with addressing uncertain variables that exist in uncontrolled settings, such as pose, expression, illumination and makeup. Facial recognition models used for practical applications usually require a large manually labelled facial recognition dataset and a low-efficiency method undermines the accuracy rate of facial recognition technologies. In the Paper, iQIYI's technical team puts forward an innovation method that entails an optimization model for unlabeled data, which helps expand the facial recognition database in a short period of time and helps effectively improve the accuracy rate of facial recognition technologies in uncontrolled settings.

At ICCV this year, iQIYI, along with Imperial College London, DeepGlint and InfiniVision, held a lightweight face recognition competition. During the competition, iQIYI released its new iQIYI-VID-FACE dataset which composes over 10,000 celebrities, 6.3 million photos and over hundreds of thousands of film and TV video clips or frame images. (For more details on the dataset, please see: http://challenge.ai.iqiyi.com/data-cluster)

In recent years, iQIYI's achievements in technical research and development have been repeatedly recognized by key international academic organizations. For example, iQIYI took the first place at Emotion Recognition in the Wild Challenge ("EmotiW2016") organized by the International Customer Management Institute ("ICMI") and published its paper titled Video-based Emotion Recognition Using CNN-RNN and C3D Hybrid Networks. In addition, iQIYI collaborated with ACM International Conference on Multimedia ("ACM MM") to hold the 2019 Celebrity Video Identification Challenge. The Challenge was participated by iQIYI along with top foreign and domestic universities, such as Carnegie Mellon University, University College London, University of Exeter, Tsinghua University, Peking University, as well as well-known companies, such as Baidu, ZTE, JD.com, Meitu, and Nvidia, achieving the precision of the recognition technology by 91.14%.

"A subject of increasing importance in AI research, multimodal person identification technology is a top priority for iQIYI researchers," said Liu Junhui, iQIYI's Senior Director. "Going forward, iQIYI will continue exploring the endless possibilities of entertainment experience by breaking through technological boundaries."

As technological advancements continue to shape the future of the entertainment industry, user experience is key to establishing one's place as an industry leader. Going forward, iQIYI will continue to push the boundaries of innovation to provide its users with optimal entertainment experiences and carry out its vision to become a technology-driven entertainment giant.