Detection, Alignment, and Recognition
October 17 at ECCV, 2008

Call for Papers

NOTE!!! Deadline for submissions extended to July 30!

Submission Instructions: Please email submissions to Erik Learned-Miller (elm@cs.umass.edu) with a subject line of "ECCV Workshop Submission" by midnight, Pacific Standard Time, of the due date. All submissions will be acknowledged via a reply email. If you do not receive a reply email within 24 hours, please inquire about the status of your submission.

Electronic submission extended to: July 30, 2008.
Decisions: August 15, 2008.
Camera ready due: August 25, 2008

We solicit contributions in two categories.

Category A: Novel methods in Detection, Alignment, and Recognition

Papers in this category should present novel scientific contributions in the detection, alignment, or recognition of faces. We are particularly interested in the domain of unconstrained faces in which faces are not presented in a laboratory controlled setting. We encourage authors to show their results on the LFW database, although this is not essential for publication. We are also interested in relationships among detection, alignment and recognition. For example, how can recognition algorithms be used to improve detection performance? Or how do various alignment algorithms effect standard recognition algorithms? We are also interested in the use of hidden variable models, random field models, and other probabilistic models for solving any of these problems. Methods that incorporate an unsupervised, semi-supervised, or transfer learning method are also solicited.

Category B: Unconstrained Face Recognition Challenge.
The goal of these submissions is to compare algorithms for the unconstrained face recognition problem, and should present results on the Labeled Faces in the Wild database . Authors may submit either a short paper or a regular paper in this Category.

For short papers (two pages or less), the authors need only include face recognition results, as described below. These results will be summarized and described by the organizers during the workshop. Authors may give a short description of their methods or refer to other publications which give the details of the algorithms used. Short papers will not appear as separate publications in the workshop proceedings, but will be described collectively in a single summary article describing results on the challenge.

For regular papers (of standard ECCV format and length), authors should fully describe algorithms so that the code can be recreated by others. If accepted, these papers will be included in the proceedings of the workshop, but the authors may or may not be allocated an oral presentation, depending upon time availability. Papers that are submitted both to the main ECCV conference and to the workshop will be considered.

Details of the LFW database, including formats of data, and organization of training, validation, and testing components, are described in the LFW technical report. It is not essential that an algorithm achieve state-of-the-art performance in order to be published at the workshop, although performance of the algorithm will be an important criterion in establishing the quality of the work. Papers that do not achieve state-of-the-art results may be published in the workshop proceedings depending upon the novelty of the proposed methods, but may not be allocated a talk, since the workshop time is limited. Such contributions will be summarized in a presentation on the overall Unconstrained Face Recognition Challenge.

Performance Reporting. Every paper in Category B should report the estimated mean accuracy and the standard error of the mean, as defined on page 7 of the LFW technical report. Users should report results for image-restricted training (described in Section IV.A. of the technical report), but may also report results for unrestricted training (Section IV.B.) if desired.

We also encourage authors to submit with their papers data files of results that allow the creation of Receiver Operating Characteristic (ROC) curves. For information on the format of ROC data files, and to create ROC curves, refer to the instructions on the Results Page of the LFW database, which is here.

Format of papers

For both Categories of papers, please use the standard ECCV template for paper submissions, including the short papers. Submissions should NOT be anonymous. Papers longer than 14 pages will be returned without review. There is no minimum paper length.