🤖 AI Summary
This work addresses the challenge of few-shot segmentation in remote sensing imagery, where scarce annotations and intra-class multimodality hinder performance. The authors propose a training-free segmentation method that leverages frozen DINOv3 self-supervised features to model foreground and background as point cloud distributions on the unit hypersphere. Departing from conventional single-prototype representations, the approach introduces a non-parametric density ratio for the first time and employs a Bayesian decision threshold for pixel-wise classification. All hyperparameters—including kernel bandwidth and spatial gating—are derived directly from the support set, enabling the decision boundary to automatically sharpen as more reference samples are provided. Evaluated across 17 remote sensing benchmarks, the method significantly outperforms existing techniques, achieving a 5.6% absolute gain in single-shot mIoU, with consistent performance improvements as the support set size increases, all while maintaining a lightweight model architecture.
📝 Abstract
Few-shot segmentation asks a model to delineate a target class in a query image from only a handful of annotated examples, a setting most acute in remote sensing, where labels are scarce and the imagery departs sharply from the natural images on which vision backbones are pretrained. Prevailing approaches either train a segmenter on labelled episodes, which raises accuracy within the training distribution but binds the model to it, or reduce each class to a lossy summary of frozen features, a single prototype, a few cluster prototypes, or a discrete clustering, none of which preserves the internal structure of a multimodal class. We argue that a class is better described by a distribution than by a point, and that frozen self-supervised features already carry enough structure to estimate that distribution directly. We introduce FROST, a training-free few-shot segmenter that treats the reference foreground and background as two point clouds on the unit sphere of frozen DINOv3 features and labels each query token by a nonparametric density ratio, with a threshold the Bayes rule fixes at zero under equal priors. Because the variance of a density estimate shrinks as its sample grows, the decision sharpens as references accumulate, and every remaining quantity from the kernel bandwidth to the spatial gate is read from the support set rather than tuned. We develop FROST for overhead imagery, where a class is typically a scatter of many small and dissimilar instances that a density tracks but a lossy summary blurs. Across seventeen remote-sensing benchmarks FROST surpasses both training-free and learning-based methods, leading by 5.6 mIoU from a single annotated example and widening its lead as the support set grows, all while remaining among the smallest models compared. Code is available at https://github.com/jhpark-ai/FROST.