S4OD: Semi-Supervised learning for Single-Stage Object Detection

Yueming Zhang,Xingxu Yao,Chao Liu,Feng Chen,Xiaolin Song,Tengfei Xing,Runbo Hu,Hua Chai,Pengfei Xu,Guoshan Zhang

DOI: https://doi.org/10.48550/arXiv.2204.04492

2022-04-09

Abstract:Single-stage detectors suffer from extreme foreground-background class imbalance, while two-stage detectors do not. Therefore, in semi-supervised object detection, two-stage detectors can deliver remarkable performance by only selecting high-quality pseudo labels based on classification scores. However, directly applying this strategy to single-stage detectors would aggravate the class imbalance with fewer positive samples. Thus, single-stage detectors have to consider both quality and quantity of pseudo labels simultaneously. In this paper, we design a dynamic self-adaptive threshold (DSAT) strategy in classification branch, which can automatically select pseudo labels to achieve an optimal trade-off between quality and quantity. Besides, to assess the regression quality of pseudo labels in single-stage detectors, we propose a module to compute the regression uncertainty of boxes based on Non-Maximum Suppression. By leveraging only 10% labeled data from COCO, our method achieves 35.0% AP on anchor-free detector (FCOS) and 32.9% on anchor-based detector (RetinaNet).

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the class imbalance between positive and negative samples encountered when applying semi - supervised learning in single - stage object detectors. Specifically, single - stage detectors directly perform dense predictions at each scale, resulting in only a small number of predictions being assigned as positive samples, thus causing a serious imbalance between positive and negative samples. This imbalance will limit the performance of single - stage detectors in semi - supervised learning. Therefore, the paper proposes a Dynamic Self - Adaptive Threshold (DSAT) strategy to automatically select pseudo - labels in order to achieve the optimal balance between the quality and quantity of pseudo - labels. In addition, in order to evaluate the regression quality of pseudo - labels in single - stage detectors, the paper also proposes an uncertainty calculation module (NMS - UNC) based on Non - Maximum Suppression (NMS). Through these methods, the paper aims to improve the performance of single - stage object detectors in the semi - supervised learning framework.