An Attempt to Build Object Detection Models by Reusing Parts

Current state-of-the-art object detection methods, like the Deformable Part Model(DPM)[3] method, have achieved remarkable performance in detecting objects. However, the annotation collecting process is very costly and it won’t scale as the number of categories increases. In this work, we make use of the information provided by part models of already trained deformable part models to decrease the amount of training data required for building good models. For easy categories like cars, aeroplanes and bicycles, we achieved decent qualitative results and quantitative results.