Knowledge-Guided Object Discovery with Acquired Deep Impressions

We present a framework called Acquired Deep Impressions (ADI) which continuously learns knowledge of objects as ``impressions'' for compositional scene understanding. In this framework, the model first acquires from images containing single object in supervised manner, and then continues to learn novel multi-object may contain that have not been seen before without any further supervision, under guidance learned humans do. By memorizing impressions into parameters neural networks applying generative replay strategy, can be reused generate with pseudo-annotations turn assist learning scenes. The proposed ADI focuses on acquisition utilization knowledge, is complementary existing deep models representation. adapt base make it fall within conduct experiments two types datasets. Empirical results suggest able effectively utilize acquired improve decomposition performance.