Alibaba iDST’s Winning Strategy in ACM MM2017 Large-Scale Video Classification

Recently, the LSVC (Large‑Scale Video Classification) competition at ACM MM2017 announced its winners, with Alibaba’s iDST (Institute of Data Science & Technologies) team taking first place thanks to a highly accurate algorithm.

ACM (Association for Computing Machinery) is a global professional organization for computing professionals, founded in 1947. ACM MM is a premier multimedia conference, and the LSVC contest is a demanding academic competition within it.

Alibaba also participated as a platinum sponsor at the ACM MM2017 conference held on October 23 in California, showcasing its multimedia technologies.

High‑Difficulty Contest in Large‑Video Domain: Alibaba Wins

LSVC targets video‑analysis researchers, testing large‑scale video data processing and classification performance. Over 20 top teams entered, with the competition providing about 62,000 uncropped videos covering 500 categories and pre‑extracted features for training, 15,000 videos for validation, and more than 78,000 videos with ground‑truth for testing.

The contest used mean Average Precision (mAP) as the evaluation metric. Alibaba’s iDST team won by a margin of 0.366 percentage points over the runner‑up.

In the competition, Alibaba leveraged Alibaba Cloud’s big‑data service ODPS to extract eight different modality features—object, scene, action, audio, etc.—pre‑trained on various databases. Experiments showed that rich multimodal information is crucial for large‑scale video classification; the best single‑model fusion combined six modalities and achieved 0.8485 mAP on the validation set. The algorithm has been deployed on the video service platform VENUS (Video Analysis and Understanding System) for video‑tag extraction tasks.

AI Achievements and Future Plans

Alibaba’s strong performance in LSVC reflects its leading position in large‑video processing and AI. At this year’s International Joint Conference on Artificial Intelligence (IJCAI), Alibaba had 11 papers accepted, and its iDST and AI LAB teams had multiple papers at CVPR.

In March, Alibaba launched the NASA program to build a powerful independent R&D institute over the next 20 years, establishing iDST, AI Labs, and the global research project “AIR” to advance foundational and breakthrough computer‑science research.