Zero-Shot Object Detection with Multi-label Context
Yongxian Wei, Yong Ma · Proceedings/Proceedings of the ... International Conference on Software Engineering and Knowledge Engineering · 2022
Zero-shot detection (ZSD) , the problem of object detection when training and test objects are disjoint, i.e. no training examples of the target classes are available.ZSD increasingly gains importance for large scale applications because collecting and labeling sufficient data is extremely hard.In this paper, inspired from human cognitive experience, we propose a simple but effective Multi-label Context (MLC) framework to facilitate the detection ability for both seen and unseen objects by mining contextual cues.We design a multilabel classifier which leverages the holistic image-level context to learn object-level concepts.Then, novel RoI features are generated by exploiting context information beneath both whole images and interested regions.Moreover, background dynamic generator (BDG) can reduce the confusion between background and unseen classes.Our extensive experiments show that MLC outperforms the current state-of-the-art methods on MS-COCO.