Strong data augmentation is a fundamental component of state-of-the-art mean teacher-based Source-Free domain adaptive Object Detection (SFOD) methods, enabling consistency-based self-supervised optimization along with weak augmentation. However, our theoretical analysis and empirical observations reveal a critical limitation: strong augmentation can inadvertently erase class-relevant components, leading to artificial inter-category confusion. To address this issue, we introduce Weak-to-strong Semantics Compensation (WSCo), a novel remedy that leverages weakly augmented images, which preserve full semantics, as anchors to enrich the feature space of their strongly augmented counterparts. Essentially, this compensates for the class-relevant semantics that may be lost during strong augmentation on the fly. Notably, WSCo can be implemented as a generic plug-in, easily integrable with any existing SFOD pipelines. Extensive experiments validate the negative impact of strong augmentation on detection performance, and the effectiveness of WSCo in enhancing the performance of previous detection models on standard benchmarks. Our code is available at https://github.com/tntek/source-free-domain-adaptive-object-detection.
•We theoretically reveal that strong augmentation in SFOD induces artificial inter-category confusion.•We propose WSCo to recover class-critical information from weakly augmented images.•Experiments show that pluggable WSCo significantly boosts SFOD model performance on benchmarks.
- Source-Free Domain Adaptive Object Detection with semantics compensation
- Song Tang - Institute of Machine Intelligence, University of Shanghai for Science and Technology, Shanghai, 200093, ChinaJiuzheng Yang - Institute of Machine Intelligence, University of Shanghai for Science and Technology, Shanghai, 200093, ChinaMao Ye - University of Electronic Science and Technology of ChinaBoyu Wang - University of LondonYan Gan - Chongqing UniversityXiatian Zhu - Surrey Institute for People-Centred Artificial Intelligence, and Centre for Vision, Speech and Signal Processing (CVSSP), University of Surrey, Guildford, UK
- Pattern recognition, Vol.179, p.113850
- Elsevier Ltd; London
- 13
- 30/04/2026
- 11/2026
- 23/04/2026
- National Natural Science Foundation of China: 62476169, 62206168, 62276048 UKRI-AHRC CoSTAR National Lab for Creative Industries Research and Development: AH/Y001060/1 German Research Foundation: GZC20233323 Postdoctoral Fellowship Program of CPSF, China: GZC20233323
This work is partly funded by the National Natural Science Foundation of China (62476169, 62206168, 62276048) ; the UKRI-AHRC CoSTAR National Lab for Creative Industries Research and Development (AH/Y001060/1) ; the German Research Foundation and National Natural Science Foundation of China in project Crossmodal Learning under contract Son-derforschungsbereich Transregio 169; the Postdoctoral Fellowship Program of CPSF, China (GZC20233323) .
- 991128105802346; WOS:001761051100001
- School of Computer Science & Electronic Engineering
- English
- Journal article