Large-Scale Unsupervised Object Discovery

Existing approaches to unsupervised object discovery (UOD) do not scale up to datasets without approximations that compromise their performance. We a novel formulation of UOD as a ranking problem, amenable to the of distributed methods available for eigenvalue problems and link. Through the use of self-supervised features, we also demonstrate the effective fully unsupervised pipeline for UOD. Extensive experiments on and OpenImages show that, in the single-object discovery setting where a prominent object is sought in each image, the proposed LOD (Large-scale Discovery) approach is on par with, or better than the state of the art medium-scale datasets (up to 120K images), and over 37% better than the other algorithms capable of scaling up to 1.7M images. In the multi-object setting where multiple objects are sought in each image, the proposed is over 14% better in average precision (AP) than all other methods for ranging from 20K to 1.7M images. Using self-supervised features, we show that the proposed method obtains state-of-the-art UOD performance on. Our code is publicly available at https://github.com/huyvvo/LOD.