Feature Fusion One-Stage Visual Affordance Detector

摘要

For a robot to be able to autonomously interact with its environment knowing only the type and location of the objects is not enough. A robot should also know what is the use of each object to be able to benefit from them. This led to the introduction of affordance detection as a crucial area of research in computer vision and robot perception in particular. In this work, a new one-stage affordance detection framework is introduced that utilizes features fusion to improve visual affordance detection performance. We used the famous UMD RGB-D Part Affordance dataset to evaluate our models and achieved comparable results to the baseline one-stage method. Extensive results are also presented to show the benefit of our method.