PixCon：基礎模型半監督分割中的乾淨正樣本對比學習

摘要

半監督語義分割（SSSS）長久以來圍繞一個核心問題打轉：該信任哪些偽標籤？並透過日益精細的信心過濾來回答。基礎模型骨幹改變了格局：使用 DINOv2 教師時，嚴格閾值已能保留經測量達 98% 純淨度的偽標籤集，因此殘留的準確率並非來自過濾器，而是來自類別如何結構化嵌入空間。我們提出 PixCon，一個乾淨正例像素對比框架。PixCon 維護每個類別的記憶庫，僅收納學生模型已正確分類的標註像素，從構造上保證無污染的正集（ρ_F=0），不同於先前基於信心過濾偽標籤的對比式 SSSS 記憶庫（如 ReCo、U^2PL）。PixCon 在一致性骨架之上採用單分支架構，無需推論階段額外參數，也不需特定於記憶庫的閾值。對監督式 InfoNCE 梯度的一階分析解釋了污染為何有害：其假正項以 ρ_F/(1-ρ_F) 比例增長，我們實際測量此值（在 Pascal 上為 0.018，在 ADE20K 上為 0.106），而非僅假設。在 Pascal VOC、Cityscapes 與 ADE20K 上，PixCon 在使用計算匹配的單一開關協議中，匹配或超越了基於強勁 DINOv2 的 UniMatch V2 基線：它提升了 Pascal-1/8 的每個種子（每種子約 +0.2 mIoU），其三種子平均值達到 87.90，即已發表的 UniMatch V2-B 數據。由於在基礎模型教師下污染已罕見，我們的分析表明，ρ_F=0 的保證主要作為教師弱化時的穩健性，而準確率提升來自更乾淨的正例監督，這使得乾淨正例對比成為基礎模型 SSSS 穩健且低成本的預設方案。

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.