Abstract:Source-Free domain adaptive Object Detection (SFOD) is a promising strategy for deploying trained detectors to new, unlabeled domains without accessing source data, addressing significant concerns around data privacy and efficiency. Most SFOD methods leverage a Mean-Teacher (MT) self-training paradigm relying heavily on High-confidence Pseudo Labels (HPL). However, these HPL often overlook small instances that undergo significant appearance changes with domain shifts. Additionally, HPL ignore instances with low confidence due to the scarcity of training samples, resulting in biased adaptation toward familiar instances from the source domain. To address this limitation, we introduce the Low-confidence Pseudo Label Distillation (LPLD) loss within the Mean-Teacher based SFOD framework. This novel approach is designed to leverage the proposals from Region Proposal Network (RPN), which potentially encompasses hard-to-detect objects in unfamiliar domains. Initially, we extract HPL using a standard pseudo-labeling technique and mine a set of Low-confidence Pseudo Labels (LPL) from proposals generated by RPN, leaving those that do not overlap significantly with HPL. These LPL are further refined by leveraging class-relation information and reducing the effect of inherent noise for the LPLD loss calculation. Furthermore, we use feature distance to adaptively weight the LPLD loss to focus on LPL containing a larger foreground area. Our method outperforms previous SFOD methods on four cross-domain object detection benchmarks. Extensive experiments demonstrate that our LPLD loss leads to effective adaptation by reducing false negatives and facilitating the use of domain-invariant knowledge from the source model. Code is available at <a class="link-external link-https" href="https://github.com/junia3/LPLD" rel="external noopener nofollow">this https URL</a>.

Gradient Adjusted and Weight Rectified Mean Teacher for Source-Free Object Detection

Balanced Teacher for Source-free Object Detection

Dynamic Retraining-Updating Mean Teacher for Source-Free Object Detection

Simplifying Source-Free Domain Adaptation for Object Detection: Effective Self-Training Strategies and Performance Insights

Periodically Exchange Teacher-Student for Source-Free Object Detection

Rethinking Weak-to-Strong Augmentation in Source-Free Domain Adaptive Object Detection

Adversarial Alignment for Source Free Object Detection

Source-Free Domain Adaptation for YOLO Object Detection

Source-free domain adaptive object detection based on pseudo-supervised mean teacher

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

Exploiting Low-confidence Pseudo-labels for Source-free Object Detection

Enhancing Source-Free Domain Adaptive Object Detection with Low-confidence Pseudo Label Distillation

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

DTG-SSOD: Dense Teacher Guidance for Semi-Supervised Object Detection

Contrastive Mean Teacher for Domain Adaptive Object Detectors

Calibrated Teacher for Sparsely Annotated Object Detection

Weakly Supervised Few-Shot Object Detection with DETR

Adapting Object Size Variance and Class Imbalance for Semi-supervised Object Detection

Multi-level domain perturbation for source-free object detection in remote sensing images

Training-based Model Refinement and Representation Disagreement for Semi-Supervised Object Detection

Multi-source-free Domain Adaptive Object Detection