PixCon: 干净正样本对比学习用于基础模型半监督分割

摘要

半监督语义分割（SSSS）长期以来围绕着一个核心问题：哪些伪标签值得信任？并一直通过日益精细的置信度过滤来回答这个问题。基础骨干网络的引入改变了这一格局：使用DINOv2教师模型时，严格的阈值即可保留测得的98%纯净度伪标签集，因此剩余精度并不依赖于过滤机制，而在于嵌入空间如何按类别结构组织。我们提出PixCon，一种基于干净正样本的像素对比学习框架。PixCon维护一个每类记忆库，仅接纳学生模型已正确分类的有标签像素，通过构造保证了无污染的正样本集（ρ_F=0），这与先前基于置信度过滤伪标签的对比SSSS记忆库（如ReCo、U^2PL）不同。PixCon仅需在一致性骨干网络上添加单一分支，不增加推理阶段参数，也无需记忆库专属阈值。对监督InfoNCE梯度的一阶分析解释了污染为何有害：其假正项按ρ_F/(1-ρ_F)规模增长，我们在Pascal上测得0.018、在ADE20K上测得0.106，而非简单假设。在Pascal VOC、Cityscapes和ADE20K数据集上，PixCon在计算量匹配的“单次开关”协议下，匹配或超越了基于DINOv2的强基线UniMatch V2：它改进了Pascal-1/8每个种子的结果（每种子约+0.2 mIoU），其三种子均值达到87.90，与已发表的UniMatch V2-B指标持平。由于基础模型教师下污染已属罕见，我们的分析表明，ρ_F=0保证主要作为教师模型变弱时的鲁棒性机制，而精度提升源自更干净的正监督。这使得干净正样本对比成为基础模型SSSS中一种鲁棒、低成本的默认选择。

English

Semi-supervised semantic segmentation (SSSS) has long turned on one question, which pseudo-labels to trust, and answered it with ever more careful confidence filtering. Foundation backbones change the regime: with a DINOv2 teacher a strict threshold already retains a measured 98%-clean pseudo-label set, so the accuracy that remains lives not in the filter but in how the embedding space is structured by class. We propose PixCon, a clean-positive pixel-contrastive framework. PixCon maintains a per-class memory bank that admits only labeled pixels the student already classifies correctly, guaranteeing a contamination-free positive set (ρ_F=0) by construction, unlike prior contrastive SSSS banks (ReCo, U^2PL) built from confidence-filtered pseudo-labels. It is a single branch over a consistency backbone, adds no inference-time parameters, and needs no bank-specific threshold. A first-order analysis of the supervised-InfoNCE gradient explains why contamination hurts: its false-positive term scales as ρ_F/(1-ρ_F), which we measure (0.018 on Pascal, 0.106 on ADE20K) rather than assume. Across Pascal VOC, Cityscapes, and ADE20K, PixCon matches or improves a strong DINOv2-based UniMatch V2 baseline in a compute-matched one-switch protocol: it improves every Pascal-1/8 seed (a per-seed gain of about +0.2 mIoU) and its three-seed mean reaches 87.90, the published UniMatch V2-B figure. Because contamination is already rare under foundation-model teachers, our analysis indicates the ρ_F=0 guarantee acts chiefly as robustness as teachers weaken, while the accuracy gain comes from cleaner positive supervision, making clean-positive contrast a robust, low-cost default for foundation-model SSSS.