---
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](https://serre-lab.clps.brown.edu/)

## 📁 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).