Abstract. Current two-stage object detectors, which consists of a region proposal stage and a refifinement stage, may produce unreliable results due to ill-localized proposed regions. To address this problem, we propose a context refifinement algorithm that explores rich contextual information to better refifine each proposed region. In particular, we fifirst identify neighboring regions that may contain useful contexts and then perform refifinement based on the extracted and unifified contextual information. In practice, our method effffectively improves the quality of the fifinal detection results as well as region proposals. Empirical studies show that context refifinement yields substantial and consistent improvements over difffferent baseline detectors. Moreover, the proposed algorithm brings around 3% performance gain on PASCAL VOC benchmark and around 6% gain on MS COCO benchmark respectively