CVPR2021

Scene Essence

Jiayan Qiu, Yiding Yang, Xinchao Wang, Dacheng Tao

摘要

Figure 1 : Given an input image of a hotel room (a), we detect its scene objects in (b) and learn to identify the Scene Essence that comprises a collection of essential elements for recognizing the scene, as labeled by the yellow bounding boxes. The image with essential elements preserved but minor ones inpainted are shown in (c), which, still, would be visually recognized as a hotel room. Should we further wipe off elements from the Scene Essence, in this case the bed, the scene will be interpreted as a living room.