Training object detectors from few weakly-labeled and many unlabeled images. (December 2021)

• Record Type: Journal Article
• Authors: Yang, Zhaohui; Shi, Miaojing; Xu, Chao; Ferrari, Vittorio; Avrithis, Yannis
• Is Part Of: Pattern recognition. Volume 120(2021)
• Publication Date: 2021-12
• Subjects: Object detection -- Weakly-supervised learning -- Semi-supervised learning -- Unlabelled set; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2021.108164
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: A novel method to train detector by few weakly-labeled images and lots of unlabeled images. The features extracted from the labeled images by a pretrained classifier are used to label unsupervised images. The proposed approaches result in competitive performance with state-of-the-art weakly supervised object detectors. Abstract: Weakly-supervised object detection attempts to limit the amount of supervision by dispensing the need for bounding boxes, but still assumes image-level labels on the entire training set. In this work, we study the problem of training an object detector from one or few images with image-level labels and a larger set of completely unlabeled images. This is an extreme case of semi-supervised learning where the labeled data are not enough to bootstrap the learning of a detector. Our solution is to train a weakly-supervised student detector model from image-level pseudo-labels generated on the unlabeled set by a teacher classifier model, bootstrapped by region-level similarities to labeled images. Building upon the recent representative weakly-supervised pipeline PCL [1], our method can use more unlabeled images to achieve performance competitive or superior to many recent weakly-supervised detection solutions. Code will be made available at https://github.com/zhaohui-yang/NSOD.