Online Pseudo-Label Unified Object Detection for Multiple Datasets Training

TL;DR

This paper introduces an online pseudo-labeling approach for unified object detection across multiple datasets, addressing missing annotations and improving accuracy through a periodically updated teacher model and category-specific box regression.

Contribution

It proposes a novel online pseudo-labeling scheme with a teacher model and category-specific box regression for improved multi-dataset object detection.

Findings

Achieves higher accuracy than SOTA methods on COCO, Object365, and OpenImages.

Effectively addresses missing annotations in cross-dataset training.

Enhances recall rate of RPN with a pseudo-label RPN head.

Abstract

The Unified Object Detection (UOD) task aims to achieve object detection of all merged categories through training on multiple datasets, and is of great significance in comprehensive object detection scenarios. In this paper, we conduct a thorough analysis of the cross datasets missing annotations issue, and propose an Online Pseudo-Label Unified Object Detection scheme. Our method uses a periodically updated teacher model to generate pseudo-labels for the unlabelled objects in each sub-dataset. This periodical update strategy could better ensure that the accuracy of the teacher model reaches the local maxima and maximized the quality of pseudo-labels. In addition, we survey the influence of overlapped region proposals on the accuracy of box regression. We propose a category specific box regression and a pseudo-label RPN head to improve the recall rate of the Region Proposal Network (PRN). Our experimental results on common used benchmarks (\eg COCO, Object365 and OpenImages) indicates that our online pseudo-label UOD method achieves higher accuracy than existing SOTA methods.