Improving Open-Set Semi-Supervised Learning with Self-Supervision

Abstract

Open-set semi-supervised learning (OSSL) is a realistic setting of semi-supervised learning where the unlabeled training set contains classes that are not present in the labeled set. Many existing OSSL methods assume that these out-of-distribution data are harmful and put effort into excluding data from unknown classes from the training objective. In contrast, we propose an OSSL framework that facilitates learning from all unlabeled data through self-supervision. Additionally, we utilize an energy-based score to accurately recognize data belonging to the known classes, making our method well-suited for handling uncurated data in deployment. We show through extensive experimental evaluations on several datasets that our method shows overall unmatched robustness and performance in terms of closed-set accuracy and open-set recognition compared with state-of-the-art for OSSL. Our code will be released upon publication.

Feature Alignment FPN for Oriented Object Detection in Remote Sensing Images

“…Prior works on open-set semi-supervised learning [48,49,50,51,52,53,54,55,56] have primarily focused on image classification tasks. For example, MTC [49] utilizes a joint optimization framework to estimate the OOD score of unlabeled images, which is achieved by alternately updating network parameters and estimated scores.…”

“…OpenMatch [50] applies consistency regularization on a one-vs-all classifier, which serves as an OOD detector to filter the OOD samples during semi-supervised learning. Wallin et al [54] propose an OSSL framework that facilitates learning from all unlabeled data through self-supervision and utilizes an energy-based score to accurately recognize data belonging to the known classes. Despite these promising results, OSSL for object detection tasks is more challenging than image classification tasks because one image typically contains more instances.…”