Results and Winning Solutions of the 2019 CCF Big Data & Computing Intelligence Contest – Video Copyright Detection Track

The 2019 CCF Big Data & Computing Intelligence Contest (CCF BDCI) finals and award ceremony were successfully concluded in Zhengzhou. The competition, which focuses on algorithms, applications, and systems challenges in big data and artificial intelligence, attracted participants from 25 countries, with a total of 25,045 teams representing 1,282 enterprises and 1,215 universities, involving 28,269 individuals.

During the award ceremony, iQIYI announced the results of the "Video Copyright Detection Algorithm" track. As a leading entertainment video platform, iQIYI has been actively exploring copyright protection technologies and hopes the contest will promote the introduction and iteration of such technologies across the industry.

After three months of online competition, 705 teams took part, including teams from top universities such as Beihang University, South China University of Technology, University of Chinese Academy of Sciences, Renmin University of China, Xidian University, Northeastern University, Nanjing University, Tsinghua University, and many enterprises.

TOP1 – Hengyang Data Team members: Li Chao, Liu Zhou, Chen Longsen, Huang Yuxiang. Organization: Hengyang Data. Solution: The team used a deep convolutional neural network (VGG16) to extract image feature vectors, performed approximate nearest‑neighbor search to find similar frames, and then matched the temporal sequence of frames to locate the start and end times with a precision of three seconds. They also employed an unsupervised Gaussian_R‑MAC aggregation method to improve robustness to large transformations.

TOP2 – Boyun Vision Team members: Xie Zhangxiang, Lou Yihang, Bai Yan, Chen Jie, Zhang Zhenbin. Organization: Boyun Vision. Solution: The team proposed a video copyright detection method based on CDVS (Compact Descriptor for Visual Search) and CDVA (Compact Descriptor for Video Analysis), achieving results comparable to the champion.

TOP3 – Xiao Jia’s Lao Liang Team member: Liang Yanjun. Organization: Beijing University of Science and Technology. Solution: The approach consists of three parts – feature extraction, infringing video retrieval, and infringing segment localization. It introduces a QAGS (Query‑Based Asymmetric Gaussian Skipping) time‑alignment scheme that improves precision without relying on machine‑learning‑based training.

TOP4 – Hulu Brothers Team members: Liu Yuzhong, Shi Jia, Yang Ye, Chen Jianqiu, Miao Shilei. Organization: JD.com / University of California, Berkeley / University of New South Wales / University of Melbourne. Solution: The team combined local and global features and optimized the temporal alignment algorithm for the dataset.

TOP5 – "Mean Beauty 98, Variance 100" Team members: Chen Zhiyuan, Yu Fan, Jiang Shilu. Organization: Beihang University. Solution: The team designed a frame‑sampling and unsupervised learning algorithm tailored to the dataset, achieving high efficiency without any machine‑learning‑based training; deep models were only used for feature extraction.

iQIYI’s "Video Copyright Detection Algorithm" challenge evaluated two dimensions: feature extraction and feature retrieval. The contest accelerated the deployment of innovative solutions and fostered deep communication between talent and enterprises. iQIYI’s director, Chen He, highlighted that the emerging ideas will support iQIYI’s copyright review and content protection services, and the technologies will be applied to short‑video and long‑video association, as well as copyright detection services.

Experts noted that the competition not only focused on performance but also on the novelty and generalizability of the technical solutions, which could be extended to other application scenarios. The diverse backgrounds of participants, especially university teams, provide valuable references for industry development, and future contests will raise the bar on innovation and depth of research.