Home
Publications
Patents
[Datasets]
Contact
Teaching
my pic


This page contains information about the following datasets I, and my colleagues, created for experimenting with different visual tasks.
  1. Dataset of Human Attributes (HAT)
  2. Dataset of TV Series Face Tubes (TSFT)

Dataset of Human Attributes (HAT)

Examples of attributes in HAT dataset

The official HAT dataset page is here. It was made when I was with GREYC CNRS UMR 6072, Univesity of Caen Normandie.

HAT contains 9344 human images labelled with 27 semantic attributes. The images were crawled from Flickr and were manually annotated by three annotators for each attribute.

Send a request to me if you are interested in obtaining our Dataset of Human Attributes (HAT).

Please cite the following paper if you use HAT dataset.

Learning Discriminative Spatial Representation for Image Classification
(Oral presentation; 8% acceptance rate)
G. Sharma, F. Jurie
British Machine Vision Conference (BMVC)
Dundee, UK, Sep 2011
PDF   Abstract   Presentation Video
@inproceedings{ sharmaBMVC2011,
            title = {Learning discriminative spatial representation for image classification},
            author = {Gaurav Sharma and Frederic Jurie},
            year = {2011},
            booktitle = {Proceedings of the British Machine Vision Conference (BMVC)}
}


Dataset of TV Series Face Tubes (TSFT)

Examples of faces in TSFT dataset

TSFT dataset contains 589* manually annotated face tubes of 94 subjects in popular TV series. The dataset has diverse set of subjects (age, gender, race, sex) in challening filming conditions (home, office, bar, inside car, day, night etc.). The average tube length is 55 frames and the average face size is 121 pixels square.

Statistics of TSFT dataset

It is a highly challenging dataset as more than 32000 faces appear in many different filming conditions and unrestricted emotions. To demonstrate the challenge qualitatively the following figure shows the false positive matches found by our method (ref. below):

False positive matches by our method in TSFT dataset

Send a request to me if you are interested in obtaining our TV Series Face Tubes (TSFT) dataset.

Please cite the following paper if you use TSFT dataset.

Latent Max-margin Metric Learning for Comparing Video Face Tubes
(Best paper award)
G. Sharma, P. Perez
Workshop on Biometrics
Computer Vision and Pattern Recognition (CVPR)

Boston, MA, USA, June 2015
PDF   Abstract
@inproceedings{sharmaCVPRW2015,
            title = {Latent Max-margin Metric Learning for Comparing Video Face Tubes},
            author = {Gaurav Sharma and Patrick Perez},
            year={2015},
            booktitle={Computer Vision and Pattern Recognition (CVPR) Workshops}
}

* The paper has a typo in Table 1, it reports 569 instead of 589 tubes.