Challenge Overview
Automatic kinship recognition holds promise to an abundance of applications. For starters, in forensics– kinship is a powerful cue that would certainly narrow the search space (e.g., knowing that the “Boston Bombers” were brothers could have helped identify the suspects sooner). In short, there are many beneficiaries that could result from such technologies: whether the consumer (e.g., automatic photo library management), scholar (e.g., historic lineage & genealogical studies), data analyzer (e.g., social-media-based analysis), or investigator (e.g., cases of missing children and human trafficking– for instance, it is unlikely that a missing child found online would be in any database, however, more than likely a family member would be. Besides application- based problems, and as already hinted, kinship is a powerful cue that could serve as a face attribute capable of greatly reducing the search space in more conventional face- recognition problems.
A fair question to ask– if so useful, then why is kinship recognition technology not found or prototyped in any real-world problem? Our reasoning is two-fold:
FIW allows us to approach this relatively new and challenging problem at sizes much greater than ever before—644,000 face pairs for 11 relationship types used for kinship verification, opposed to just 2,000 for 4 types; Also, 1,000 families for family classification, opposed to just 101. In the end, we hope FIW provides a rich resource to further advance automatic facial understanding technologies as part of the broader human-computer interaction incentive.
FIW span data distributions that more appropriately mock real world scenario. We expect the larger, more complex data will pave way to more attempts to employ modern day data driven (i.e., deep) methods in ways that were not possible before.
A fair question to ask– if so useful, then why is kinship recognition technology not found or prototyped in any real-world problem? Our reasoning is two-fold:
- Existing image datasets for kinship recognition tasks are not large enough to capture and reflect the true data distributions of the families of the world.
- Kin-based relationships in the visual domain are less discriminant than other, more conventional problems of its kind (e.g., facial recognition or object classification), as there exists many hidden factors that affect the facial appearances amongst different family members.
FIW allows us to approach this relatively new and challenging problem at sizes much greater than ever before—644,000 face pairs for 11 relationship types used for kinship verification, opposed to just 2,000 for 4 types; Also, 1,000 families for family classification, opposed to just 101. In the end, we hope FIW provides a rich resource to further advance automatic facial understanding technologies as part of the broader human-computer interaction incentive.
FIW span data distributions that more appropriately mock real world scenario. We expect the larger, more complex data will pave way to more attempts to employ modern day data driven (i.e., deep) methods in ways that were not possible before.
Data Challenge Workshop
The data challenge workshop will consist of 2 tasks:
- Kinship Verification (one-vs-one)
- Family Classification (one-vs-many)
A Broader Impact
The size of FIW, along with the labels representing complex family structures of 1,000 families, make it hard to pinpoint which of the many possible directions it will lead. Improving upon the evaluation benchmarks is one obvious route, which is the focus of the data challenge portion of this workshop. However, additional task evaluations (e.g., search & retrieval), along with cross-discipline studies (e.g., nature-based) are certainly subject to being possible directions for future work. By the efforts and feedback from the participates and organizers, supported with material that will be delivered during the workshop itself, the aim is to discuss current progress, generate innovative ideas, and solidify plans for next steps (e.g., future challenges).
We anticipate that the data challenge workshop will fairly reflect current progress in our automatic kinship recognition capabilities, along with bring forth discussions for future work and collaborations with FIW. Additionally, we expect our resources (i.e., source code) will help lessen the learning curve for newcomers, while the problem will challenge the experts (i.e., attracting the experts, while enabling newcomers).
We expect the following benefits will result from the data challenge workshop:
We anticipate that the data challenge workshop will fairly reflect current progress in our automatic kinship recognition capabilities, along with bring forth discussions for future work and collaborations with FIW. Additionally, we expect our resources (i.e., source code) will help lessen the learning curve for newcomers, while the problem will challenge the experts (i.e., attracting the experts, while enabling newcomers).
We expect the following benefits will result from the data challenge workshop:
- Gain better understanding of state-of-the-art for kinship recognition technologies.
- Enable various data-driven methods (i.e., deep learning) methods to be employed.
- Find shortcomings of the evaluations for which to improve on; Also, identify the parts that work well.
- Pinpoint future directions for the FIW database, along with additional label types and data from different modalities (e.g., text captions for family photos)— resources to broaden the scope of future challenges, benchmarks, and studies.
- Broaden the list of practical uses for automatic kinship recognition technologies.
- Attract newcomers to the problem, along with challenge the experts.
- Generate connections between groups with different strengths, as FIW contains much more information than provided by the faces alone.
- Provide platform for inter-discipline studies between life sciences and machine vision communities.