Abstract: False correspondence removal remains a challenge for many feature-matching-based applications. This paper proposes a solution that revisits the local semantic context via position-attentive ...