YouTube Celebrities Face Tracking and Recognition Dataset

This dataset is released as a part of the work described in

Kim, M., Kumar, S., Pavlovic, V. & Rowley, H. (2008), "Face Tracking and Recognition with Visual Constraints in Real-World Videos", In IEEE Conf. Computer Vision and Pattern Recognition. Anchorage, AK. June 2008.

Please reference the above paper if you use this set in your work.

More details about this work, including demonstration videos, can be found on our Face Project page.

Description

The data consists of two parts: (1) video clips and (2) initialization data.

Video Clips

The dataset contains 1910 sequences of 47 subjects. All videos are encoded in MPEG4 at 25fps rate. The filenames conform to the following convention:

aaaa_bb_ccc_name.avi

where

aaaa - unique sequence ID
bb - subject clip ID
ccc - video segment ID
name - name of the subject

For example, "1906_03_007_vladimir_putin.avi" means sequence ID 1906, clip 03 (out of 3 clips from the same subject), 7th segment from the 3rd clip, and subject Vladimir Putin.

Initialization

For each video, this MATLAB file contains initial frame face bounding boxes, manually marked, represented as an affine parameter vector. The 6-dim vector contains (pos-x, pos-y, scale, rot, *,*). The last two parameters are related to a skew transform and can be ignored.

Download

The tar file of video clips can be downloaded from here. Note that this file is about 885MB in size. Initial parameters (2MB) can be downloaded from here.