Spatial reasoning for few-shot object detection. (December 2021)

• Record Type: Journal Article
• Authors: Kim, Geonuk; Jung, Hong-Gyu; Lee, Seong-Whan
• Is Part Of: Pattern recognition. Volume 120(2021)
• Publication Date: 2021-12
• Subjects: Few-shot learning -- Object detection -- Transfer learning -- Visual reasoning -- Data augmentation; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2021.108118
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: We consider few-shot object detection that requires only a few training examples to detect novel categories. Inspired by a human visual system, we propose a spatial reasoning process to detect novel categories in a context that is less considered in few-shot object detection. To overcome a few-shot environment itself, we further present a spatial data augmentation method that efficiently enhance the ability of the spatial reasoning process. The proposed method significantly outperforms existing few-shot object detectors on the widely used PASCAL VOC and MS COCO datasets. Abstract: Although modern object detectors rely heavily on a significant amount of training data, humans can easily detect novel objects using a few training examples. The mechanism of the human visual system is to interpret spatial relationships among various objects and this process enables us to exploit contextual information by considering the co-occurrence of objects. Thus, we propose a spatial reasoning framework that detects novel objects with only a few training examples in a context. We infer geometric relatedness between novel and base RoIs (Region-of-Interests) to enhance the feature representation of novel categories using an object detector well trained on base categories. We employ a graph convolutional network as the RoIs and their relatedness are defined as nodes and edges, respectively. Furthermore, we present spatial data augmentation to overcome the few-shot environment where all objects and bounding boxes in an image are resized randomly. Using the PASCAL VOC and MS COCO datasets, we demonstrate that the proposed method significantly outperforms the state-of-the-art methods and verify its efficacy through extensive ablation studies.