HMDB51: A Large Video Database for Human Motion Recognition Official
HMDB51 contains 6,849 short video clips across 51 action categories, compiled from movies and web videos to capture diverse viewpoints, camera motion, and visual conditions.
Downloads
Community Mirrors and Tools
Additional community mirrors provided for convenience.
Action Classes (51)
General facial actions
Facial actions with object manipulation
General body movements
- cartwheel, clap hands, climb, climb stairs, dive, fall on the floor, backhand flip, handstand, jump, pull up, push up, run, sit down, sit up, somersault, stand up, turn, walk, wave
Body movements with object interaction
- brush hair, catch, draw sword, dribble, golf, hit something, kick ball, pick, pour, push something, ride bike, ride horse, shoot ball, shoot bow, shoot gun, swing baseball bat, sword exercise, throw
Human–human interaction
- fencing, hug, kick someone, kiss, punch, shake hands, sword fight
Meta-Labels
Each video clip is annotated with meta-labels describing various attributes:
Viewpoint
- side_view, front_view, back_view, top_view, bottom_view
Video quality
- low_quality, medium_quality, high_quality
Illumination
- dark, medium_light, bright
Focal length
- short_focal, medium_focal, long_focal
Video Stabilization
The original HMDB51 dataset contains unstabilized videos. A stabilized version is also available, where geometric frame alignment has been applied to reduce camera motion. The unstabilized version is the official release.
Citation
5,336 citations on Google Scholar as of November 2025.
Please cite the following paper when using this dataset:
H. Kuehne, H. Jhuang, E. Garrote, T. Serre and T. Poggio. HMDB: A large video database for human motion recognition. ICCV, 2011.