Computer Vision News - January 2020
3 Summary Families In the Wild 2 1. Kin-based datasets must capture true data distributions of familial relationships as seen around the globe. 2. Recognizing kinship in faces is a challenging task, as many hidden factors affect the resemblance and shared characteristics of a given family. FIW was built with such specifications in mind. Thus, its contents closely mimic data distributions and sample sizes as would be seen in many real-world scenarios. This larger, more complex data has the potential to satisfy the capacities of modern-day data-driven methods. Furthermore, with our FIW database, we can pose challenging problems at sizes more massive than before possible. Thus, it is our intent to provide the needed resources to bridge the gap of Face Recognition technology as part of the broader HCI (Human-Computer Interaction) incentive . Introduced in2016, and further extended recently, FIW remains the largest and most comprehensive database for kinship recognition. With over 14,000 family photos of 1,000 families captured in the wild, a list of over 1M face-pairs of 11 different relationship types provides settings for the problem never before possible. Furthermore, FIW makes great strides to provide a data resource that matches the distribution of familial structure as seen in nature. The 4th RFIW calls for general papers and offers three task evaluations that are open for individuals or teams on Codalab [verification ( T1 ) ; tri-subject ( T2 ) ; search & retrieval ( T3 ) ]! All paper submissions are subject to peer-review for publication in the Proceedings on FG2020, which will be held on 18-22 May 2019 in Buenos Aires, Argentina. We expect general paper submissions on vast views and approaches to problems of kinship: including, but not limited to, generative models (e.g., predict appearances of family members), fine-grained recognition (e.g., complete family trees), and gained insights from data analysis and processing (e.g., clustering). Background Challenge & Workshop Proceedings
Made with FlippingBook
RkJQdWJsaXNoZXIy NTc3NzU=