i3DPost Multi-view Human Action Datasets

This is a webpage to distribute multi-view/3D human action/interaction datasets created in a cooperation between University of Surrey and CERTH-ITI within the i3DPost project.

All details about the datasets are described in this paper: N. Gkalelis, H. Kim, A. Hilton, N. Nikolaidis and I. Pitas, "The i3DPost multi-view and 3D human action/interaction," Proc. CVMP, pp. 159-168, 2009 PDF

In order to access the full datasets, please read this license agreement and send email Dr. Hansung Kim with the following information if you agree: Your name/affiliation, name/email of your supervisor (if you are a student)

Access to the datasets (ID/PW required)

If you get a permission to access, the following datasets are provided.


- Multi-view datasets - Synchronised/uncompressed-HD 8 view image sequences of 13 actions of 8 people (104 total)


- Sample videos

Actors/Actresses: Man1, Woman1, Man2, Man3, Woman2, Man4, Man5, Man6,


Actions: Walk, Run, Jump, Bend, Hand-wave, Jump-in-place, Sit-StandUp,

Run-fall, Walk-sit, Run-jump-walk, Handshake, Pull, Facial-expressions


- Background images for matting


- Camera calibration parameters (intrinsic and extrinsic) for 3D reconstruction


- 3D mesh models - All frames and actions of above datasets in ntri ascii format

The models were reconstructed using a global optimisation method: J. Starck and A. Hilton, Surface capture for performance based animation, IEEE Comp. Graph. Appl., 27(3), pp.2131, 2007. PDF


Centre for Vision, Speech and Signal Processing, University Of Surrey, Guildford, UK

Artificial Intelligence & Information Analysis Lab, CERTH-ITI, Thessaloniki,  GREECE

i3DPost, EU Project under Framework 7 ICT Programme

To contact us: Dr. Hansung Kim (h.kim@surrey.ac.uk)