Domain-RAG
Domain-RAG
Cross-Domain Few-Shot Object Detection (CD-FSOD) aims to detect novel objects with only a handful of labeled samples from previously unseen domains. While data augmentation and generative methods have shown promise in few-shot learning, their effectiveness for CD-FSOD remains unclear due to the need for both visual realism and domain alignment. Existing strategies, such as copy-paste augmentation and text-to-image generation, often fail to preserve the correct object category or produce backgrounds coherent with the target domain, making them non-trivial to apply directly to CD-FSOD. To address these challenges, we propose Domain-RAG, a training-free, retrieval-guided compositional image generation framework tailored for CD-FSOD. Domain-RAG consists of three stages: domain-aware background retrieval, domain-guided background generation, and foreground-background composition. Specifically, the input image is first decomposed into foreground and background regions. We then retrieve semantically and stylistically similar images to guide a generative model in synthesizing a new background, conditioned on both the original and retrieved contexts. Finally, the preserved foreground is composed with the newly generated domain-aligned background to form the generated image. Without requiring any additional supervision or training, Domain-RAG produces high-quality, domain-consistent samples across diverse tasks, including CD-FSOD, remote sensing FSOD, and camouflaged FSOD. Extensive experiments show consistent improvements over strong baselines and establish new state-of-the-art results.
We propose Domain-RAG, a novel training-free, retrieval-guided compositional generation framework that enhances support diversity by generating domain-aligned samples. To enable retrieval, we use COCO as the database $ \mathcal{D}_{\text{base}} $, serving as a gallery of candidate backgrounds. Following the core principle of “fix the foreground, adapt the background”, Domain-RAG processes each support image $ x \in \mathcal{S} $ by first decomposing it into foreground object(s) and background. The framework then proceeds through three key stages: (1) domain-aware background retrieval first obtains the inpainted background $ b_{\text{init}} $ from $ x $ and then retrieves $ G $ candidate backgrounds $ b_{\text{re}} $ from $ \mathcal{D}_{\text{base}} $ that are semantically and stylistically similar; (2) domain-guided background generation feeds each $ { b_{\text{init}}, b_{\text{re}} } $ pair into a generative model to synthesize a new domain-aligned background $ b_{\text{dom}} $; (3) foreground-background composition finally produces $ n $ new images $ x^{+} $ by compositing the preserved foreground onto each $ b_{\text{dom}} $ using a mask-guided generative model.
We evaluate Domain-RAG on multiple benchmarks, demonstrating consistent improvements.
