MIAdapt: Source-free Few-shot Domain Adaptive Object Detection for Microscopic Images

TL;DR

MIAdapt addresses domain shift in microscopic imagery under source-free and few-shot constraints by leveraging a pre-trained source detector and a small labeled target set. It introduces two key components: Resolution-aware augmentation (RAug) to balance classes while preserving spatial resolution, and Category-aware representation learning (CL) to align intra-class features and enhance inter-class separation, implemented within a Mean Teacher framework. The approach is evaluated on M5-Malaria and Raabin-WBC across 2-shot and 5-shot settings, showing substantial improvements over competitive baselines and source-data-free variants. The results suggest MIAdapt as a practical SF-FSDA solution for privacy-sensitive medical imaging, bridging the gap when source data are unavailable and target annotations are scarce; code will be released publicly.

Abstract

Existing generic unsupervised domain adaptation approaches require access to both a large labeled source dataset and a sufficient unlabeled target dataset during adaptation. However, collecting a large dataset, even if unlabeled, is a challenging and expensive endeavor, especially in medical imaging. In addition, constraints such as privacy issues can result in cases where source data is unavailable. Taking in consideration these challenges, we propose MIAdapt, an adaptive approach for Microscopic Imagery Adaptation as a solution for Source-free Few-shot Domain Adaptive Object detection (SF-FSDA). We also define two competitive baselines (1) Faster-FreeShot and (2) MT-FreeShot. Extensive experiments on the challenging M5-Malaria and Raabin-WBC datasets validate the effectiveness of MIAdapt. Without using any image from the source domain MIAdapt surpasses state-of-the-art source-free UDA (SF-UDA) methods by +21.3% mAP and few-shot domain adaptation (FSDA) approaches by +4.7% mAP on Raabin-WBC. Our code and models will be publicly available.

Figures (5)

• Figure 1: Source-free few-shot domain adaptation setting vs the most similar settings.
• Figure 2: MIAdapt: Our method enhances source free few shot adaptation by incorporating intelligent augmentation and category aware feature learning.
• Figure 3: (a) CBCP inayat2024few ignores image resolution difference within domain. (b) Ours, RAug preserves spatial resolution consistency, improves performance (Tab. \ref{['tab:ablation_RAug']})
• Figure 4: Detections missed by source model and found by our method are shown in Blue. MIAdapt enhances the RPN, better adapted to background-similar classes.
• Figure 5: False positives: red, MIAdapt-only detections: blue, common correct: green.