CVPR2025

Detecting Open World Objects via Partial Attribute Assignment

Muli Yang, Gabriel James Goenawan, Huaiyuan Qin, Kai Han, Xi Peng, Yanhua Yang, Hongyuan Zhu

摘要

We follow [S2] to solve the Partial Optimal Transport (POT) problem. In particular, the POT problem with entropic regularization in Eq. ( 6 ) can be rewritten as a Kullback-Leibler projection: min T∈Π(V,A) ⟨T, C⟩ F -ϵh(T) = ϵ min T∈Π(V,A) KL(T|D) , (A) in which D = exp -C ϵ and KL(•|•) denotes the Kullback-Leibler divergence: KL(T|D) ≜ M,N m,n=1