DSD-DA: Distillation-based Source Debiasing for Domain Adaptive Object Detection

TL;DR

This paper introduces a novel distillation-based framework for domain adaptive object detection that reduces source bias and enhances target domain performance through knowledge distillation, localization mining, and consistency strategies.

Contribution

It proposes a source debiasing distillation framework with a target-relevant localization network and a consistency strategy, addressing source bias and improving detection in target domains.

Findings

Significant performance improvements over baseline methods.

Outperforms existing alignment-based domain adaptation methods.

Effective in harmonizing classification and localization in target domain.

Abstract

Though feature-alignment based Domain Adaptive Object Detection (DAOD) methods have achieved remarkable progress, they ignore the source bias issue, i.e., the detector tends to acquire more source-specific knowledge, impeding its generalization capabilities in the target domain. Furthermore, these methods face a more formidable challenge in achieving consistent classification and localization in the target domain compared to the source domain. To overcome these challenges, we propose a novel Distillation-based Source Debiasing (DSD) framework for DAOD, which can distill domain-agnostic knowledge from a pre-trained teacher model, improving the detector's performance on both domains. In addition, we design a Target-Relevant Object Localization Network (TROLN), which can mine target-related localization information from source and target-style mixed data. Accordingly, we present a Domain-aware Consistency Enhancing (DCE) strategy, in which these information are formulated into a new localization representation to further refine classification scores in the testing stage, achieving a harmonization between classification and localization. Extensive experiments have been conducted to manifest the effectiveness of this method, which consistently improves the strong baseline by large margins, outperforming existing alignment-based works.