ConR proposes a contrastive regularizer to address deep imbalanced regression by modeling global and local label similarities in feature space.
