HMDB51 Dataset

HMDB51 is a large human action recognition dataset consisting of 6,766 video clips (.avi) distributed across 51 action categories (e.g., jumping, drinking, kicking, etc). It is commonly used as a benchmark for video classification and action recognition tasks.

📦 Dataset Source

This dataset was originally collected and published by the Serre Lab at Brown University.

🔗 Serre Lab – Brown University

📁 Contents

• 6,766 video clips
• 51 action categories
• Includes training/testing splits for benchmarking

📚 Citation

If you use this dataset, please cite:

Kuehne, H., Jhuang, H., Garrote, E., Poggio, T., & Serre, T. (2011). HMDB: A large video database for human motion recognition. In Proceedings of the IEEE International Conference on Computer Vision (ICCV).