In the past decades, facial image analysis has attracted continuous attention in computer vision, pattern recognition and machine learning areas, owing to its scientific challenges in both psychological interpretation and computational simulation, as well as its huge potentials in real-world applications. Much progress has been achieved in the last decades; however, researchers in the field also meet bafflement and challenges on the comprehensive and unbiased evaluation of the related technologies, which may prevent them from discovering the actual state of the art.
BeFIT –Benchmarking Facial Image Analysis Technologies– is an international collaborative effort on standardizing the evaluation of facial image analysis technologies. The objective is to bring together different face analysis evaluations and provide a medium for researchers to discuss about different aspects of face analysis. Sample categories are:
- Face Detection and Tracking
- Face Detection
- Face Tracking
- Face Alignment
- Facial Feature Localization
- Face Alignment
- Face Image Enhancement
- Super-resolution of Face Images
- Image Quality Enhancement
- Illumination Normalization
- 3D Face Reconstruction from 2D View(s)
- Face Recognition
- Open- / Close-set Face Identification / Watch List Screening
- Face Verification / Face Re-Identification / Face Clustering
- Video-based Face Recognition
- High-resolution Face Recognition
- 3D Face Recognition
- Face Attributes Analysis
- Age Estimation
- Gender Classification
- Ethnicity Classification
- Glasses, Mustache, Beard Detection
- Facial Expression Analysis
- Action Unit Detection
- Emotion Classification
The workshop aims at encouraging researchers to run more standardized tests, to share data and to propose new challenges related to face analysis. It is of particular interest to have challenges with realistic data, containing illumination and view variations. The workshop will accept following type of paper submissions:
- Face analysis papers as long as they follow a standard benchmark process or the authors would share their own data that they have used for their experiments.
- Papers presenting new data sets and proposing new challenges.
- Papers that discuss the usefulness or inability of existing benchmarks.
For the first point, these benchmarks can be any that have been already used by the community, such as FERET, FRGC, LFW, FDDB, etc.
Paper Formatting: Papers are limited to 10 pages. Please use the ECCV workshop author kit to format the papers.
Blind review: Reviewing will be double blind. Please see the ECCV workshop author kit for detailed explanations of how to ensure this.
Submission: Through the ECCV 2012 submission system.