Sourceless Object Detection with Transformer detection

Total: 1

Source-Free Object Detection (SFOD) enables the transfer of knowledge from a source domain to an unsupervised target domain for object detection without access to the source data. Most existing SFOD approaches are either confined to conventional object detection (OD) models like Faster R-CNN, or designed as general solutions without tailor-made adaptations for new OD architectures, particularly Detection Transformer (DETR). In this article, we present the Feature Reweighting AND Contrastive Learning Network (FRANCK), a new SFOD framework specifically designed to perform query-centric feature enhancement for DETRs. FRANCK includes four key components: (1) an Objectness Score-based Sample Reweighting (OSSR) module that calculates attention-based objectness scores on multi-scale encoder feature maps, reweighting detection loss to emphasize less recognized regions; (2) a contrastive learning module with matching-based memory bank (CMMB) that integrates multi-level features into memory banks, thereby improving class-wise contrastive learning; (3) an uncertainty-weighted query fused feature distillation (UQFD) module that improves feature distillation through prediction quality reweighting and query feature fusion; and (4) an enhanced self-training pipeline with dynamic teacher update interval (DTUI) that optimizes pseudo-label quality. By exploiting these components, FRANCK efficiently adapts a source-pretrained DETR model to a target domain with improved robustness and generalization. Extensive experiments on several widely used benchmarks demonstrate that our method achieves state-of-the-art performance, highlighting its effectiveness and compatibility with DETR-based SFOD models.