iQIYI Releases World's First Multimodal, Multi-angle Celebrity Video Dataset (iQIYI-VID) and Announces AI Competition

iQIYI (NASDAQ: IQ) recently released the world’s first multimodal, multi-angle celebrity dataset (iQIYI-VID), comprising 1,000 hours of video, 500,000 clips, and 5,000 celebrities. The dataset will be used in the iQIYI‑PRCV2018 Multimodal Video Person Recognition Challenge, providing real‑world, full‑angle video material for participants.

To ensure the competition’s authority, iQIYI invited a ‘dream team’ of renowned computer‑vision scholars as judges, including Sun Jian, chief scientist of Megvii, who emphasized that person‑recognition technology underpins many AI unicorns and that multimodal video person recognition can advance comprehensive human understanding.

The panel also includes Wang Liang, a researcher at the Chinese Academy of Sciences and a recipient of the National Outstanding Youth Science Fund, who highlighted the importance and challenge of multimodal person recognition and praised the iQIYI‑VID dataset as the largest celebrity video collection worldwide. Other judges are Shan Shiguang, Liu Wenfeng (CTO of iQIYI), Xie Danming (Vice President of iQIYI), and Wang Tao, senior scientist at iQIYI.

By early August, the competition had attracted hundreds of teams from top universities and research institutes such as Tsinghua University, Peking University, New York University, Singapore International University, Lund University (Sweden), and Tokyo Institute of Technology.

Unlike most computer‑vision contests that focus on face recognition, this is the first global competition dedicated to multimodal video person recognition, which integrates behavior, face, voice, and image cues. Real‑world video brings challenges such as varied poses, expressions, ages, lighting, resolution, makeup, and occlusion, and current technology still falls short of the accuracy required for practical applications.

Previously, research relied on public face datasets such as MegaFace (University of Washington) and LFW (UMass Amherst). iQIYI’s iQIYI‑VID is the largest video dataset to date, fully manually annotated, containing 5,000 celebrities and 1,000 hours of video across 500,000 clips.

In the video domain, AI‑driven understanding of audiovisual data enables fine‑grained emotion analysis, content recommendation, and production efficiency. iQIYI has applied person‑recognition technology to features like ‘Watch Only This Actor’ in popular series and rapid clip retrieval in post‑production of shows such as ‘The Rap of China’.

At iQIYI World Conference, CEO Gong Yu emphasized that personalized content creation and distribution will transform the entertainment industry.

Building on cloud computing, big data, and AI, iQIYI is establishing an open service platform and an ‘AI + software + hardware’ ecosystem. The AI challenge continues to explore frontier technologies and their deep integration with entertainment, driving both technical progress and industry adoption.

The competition registration ends on September 17. iQIYI will release a test set on the same day, with dynamic ranking of submissions until October 15. Final results will be announced on November 1, and awards will be presented at the PRCV2018 conference on November 23.

For more information, click ‘Read the original article’.