Large-Scale Unsupervised Object Discovery - Supplementary Material - Huy V. Vo

We include in this appendix additional information about the proposed method, including implementation details, experimental results, visualizations and proofs. See (1; 10) for more details. Since the adjacency matrix (in Q) and the PageRank matrix (in P) are very large, we divide them into chunks of consecutive rows of approximately equal size. A.3 Influence of the number of object proposals on object discovery performance Influence of the number of object proposals on performance of LOD. We also present some failure cases in Figure 1.