metadata

annotations_creators:
  - expert-generated
language:
  - und
multilingual: false
pretty_name: HMDB51
task_categories:
  - video-classification
license: other
dataset_creators:
  - Serre Lab at Brown University
paperswithcode_id: hmdb51
tags:
  - human-action
  - video
  - benchmark

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).