Annotator-dependent uncertainty-aware estimation of gait relative attributes

Allam Shehata, Yasushi Makihara, Daigo Muramatsu, Md Atiqur Rahman Ahad, Yasushi Yagi

Abstract: Highlights•End-to-end uncertainty-aware estimation framework for relative attributes.•Conversion module from continuous score distribution to discrete label distribution.•Proposed uncertainty layers to estimate gait relative attribute distribution.•Comprehensive experimental evaluation using two gait relative attribute datasets.•The state-of-the-art performance with average accuracy ranging from 3% to 20%.