Amodal Completion and Size Constancy in Natural Scenes
This work addresses the problem of improving object detection systems with accurate size and depth estimates for applications in robotics or autonomous driving, but it appears incremental as it builds on existing advances in object recognition and datasets.
The paper tackles the problem of estimating veridical object sizes and relative depths from single images, addressing challenges like occlusions and scale ambiguity, and demonstrates qualitative results on real-world scenes.
We consider the problem of enriching current object detection systems with veridical object sizes and relative depth estimates from a single image. There are several technical challenges to this, such as occlusions, lack of calibration data and the scale ambiguity between object size and distance. These have not been addressed in full generality in previous work. Here we propose to tackle these issues by building upon advances in object recognition and using recently created large-scale datasets. We first introduce the task of amodal bounding box completion, which aims to infer the the full extent of the object instances in the image. We then propose a probabilistic framework for learning category-specific object size distributions from available annotations and leverage these in conjunction with amodal completion to infer veridical sizes in novel images. Finally, we introduce a focal length prediction approach that exploits scene recognition to overcome inherent scaling ambiguities and we demonstrate qualitative results on challenging real-world scenes.