面向持续测试时分割的双曲原型学习与不确定性感知一致性
Hyperbolic Prototype Learning with Uncertainty-Aware Consistency for Continual Test-Time Segmentation
把语义分割的特征空间从欧氏空间换到 Poincaré 球做原型学习，用双曲体积指数增长换取更大的类间间隔；配合按跨视角一致性把像素分成 core/boundary 两组、分别用测地距离最小化和双曲方向一致性损失约束，是可直接迁移到其他 CTTA/域偏移分割流程的即插即用模块。
原文依据 We propose HyperProtoSeg, a hyperbolic prototypical segmentation network that learns geometrically optimal class prototypes in the Poincare ball.
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论面向语义分割的持续测试时自适应（CTTA）对于在存在持续域偏移的动态环境中部署视觉模型至关重要。现有方法往往随时间退化，因为自监督更新会放大早期的预测错误。我们将这种脆弱性归因于一个几何层面的局限：欧氏特征空间具有多项式体积增长，导致语义表示失真以及决策边界拥挤、不稳定。我们提出 HyperProtoSeg，一种双曲原型分割网络，在 Poincaré 球中学习几何最优的类别原型。借助双曲空间的指数级扩张特性，它以低失真强制实现大而均匀的类间间隔，从而得到良好分离且曲率稳定的嵌入。为实现稳健的在线自适应，我们引入双曲边界一致性自适应（HBCA），它依据跨视角一致性将像素划分为置信的“核心”集合与不确定的“边界”集合。HBCA 对置信区域施加测地距离最小化，对不确定区域施加一种新的双曲方向一致性损失，从而防止误差放大。在具有挑战性的合成到真实基准（Cityscapes - ACDC、IDD - IDD-AW、SHIFT）上的实验表明，在严重结构偏移下，HyperProtoSeg + HBCA 相较最优的 CTTA 方法平均提升（1.94%，4.02%，1.24%）。
英文原文摘要 1404 字符
Continual Test-Time Adaptation (CTTA) for semantic segmentation is vital for deploying vision models in dynamic environments with persistent domain shifts. Existing methods often degrade over time as self-supervised updates amplify early prediction errors. We attribute this fragility to a geometric limitation: Euclidean feature spaces, with polynomial volume growth, lead to distorted semantic representations and crowded, unstable decision boundaries. We propose HyperProtoSeg, a hyperbolic prototypical segmentation network that learns geometrically optimal class prototypes in the Poincare ball. Leveraging the exponential expansion of hyperbolic space, it enforces large and uniform inter-class margins with low distortion, yielding well separated and curvature-stable embeddings. For robust online adaptation, we introduce Hyperbolic Boundary Consistency Adaptation (HBCA), which partitions pixels by cross-view consistency into confident "core" and uncertain "boundary" sets. HBCA applies geodesic distance minimization for confident regions and a novel Hyperbolic Directional Consistency Loss for uncertain ones, preventing error amplification. Experiments on challenging synthetic- to-real benchmarks (Cityscapes - ACDC, IDD - IDD-AW, SHIFT) show that HyperProtoSeg + HBCA achieves an average improvement of (1.94%,4.02%,1.24%) over state-of-the-art CTTA methods under severe structural shifts.