2016/06/14 by Spyros Gidaris, Gidaris, Spyros, Nikos Komodakis +1 · 1 citation
Computer Science · #Advanced Neural Network Applications #Computer Vision and Pattern Recognition (cs.CV) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Multimodal Machine Learning Applications
paper · pdf · doi:10.48550/arxiv.1606.04446
openalex publication_date 2016/06/14 · openalex created_date 2019/07/30 · openalex updated_date 2026/08/01
The problem of computing category agnostic bounding box proposals is utilized\nas a core component in many computer vision tasks and thus has lately attracted\na lot of attention. In this work we propose a new approach to tackle this\nproblem that is based on an active strategy for generating box proposals that\nstarts from a set of seed boxes, which are uniformly distributed on the image,\nand then progressively moves its attention on the promising image areas where\nit is more likely to discover well localized bounding box proposals. We call\nour approach AttractioNet and a core component of it is a CNN-based category\nagnostic object location refinement module that is capable of yielding accurate\nand robust bounding box predictions regardless of the object category.\n We extensively evaluate our AttractioNet approach on several image datasets\n(i.e. COCO, PASCAL, ImageNet detection and NYU-Depth V2 datasets) reporting on\nall of them state-of-the-art results that surpass the previous work in the\nfield by a significant margin and also providing strong empirical evidence that\nour approach is capable to generalize to unseen categories. Furthermore, we\nevaluate our AttractioNet proposals in the context of the object detection task\nusing a VGG16-Net based detector and the achieved detection performance on COCO\nmanages to significantly surpass all other VGG16-Net based detectors while even\nbeing competitive with a heavily tuned ResNet-101 based detector. Code as well\nas box proposals computed for several datasets are available at::\nhttps://github.com/gidariss/AttractioNet.\n