Source-free Domain Adaptation for Video Object Detection Under Adverse Image Conditions

TL;DR

This paper introduces STAR-MT, a source-free domain adaptation method that enhances one-stage video object detection under adverse conditions like noise and haze, demonstrating consistent improvements in challenging real-world scenarios.

Contribution

The paper presents STAR-MT, the first source-free domain adaptation approach tailored for one-stage video object detection under adverse image conditions.

Findings

Improves detection performance on degraded videos

Effective under various adverse conditions like noise and haze

Shows consistent gains on ImageNetVOD dataset

Abstract

When deploying pre-trained video object detectors in real-world scenarios, the domain gap between training and testing data caused by adverse image conditions often leads to performance degradation. Addressing this issue becomes particularly challenging when only the pre-trained model and degraded videos are available. Although various source-free domain adaptation (SFDA) methods have been proposed for single-frame object detectors, SFDA for video object detection (VOD) remains unexplored. Moreover, most unsupervised domain adaptation works for object detection rely on two-stage detectors, while SFDA for one-stage detectors, which are more vulnerable to fine-tuning, is not well addressed in the literature. In this paper, we propose Spatial-Temporal Alternate Refinement with Mean Teacher (STAR-MT), a simple yet effective SFDA method for VOD. Specifically, we aim to improve the performance of the one-stage VOD method, YOLOV, under adverse image conditions, including noise, air turbulence, and haze. Extensive experiments on the ImageNetVOD dataset and its degraded versions demonstrate that our method consistently improves video object detection performance in challenging imaging conditions, showcasing its potential for real-world applications.