CW-BASS v2：基于基础模型教师的半监督分割中饱和度感知的伪标签选择

摘要

半监督语义分割长期以来一直围绕着一个问题展开：该信任哪些伪标签。针对当时噪声大、置信度不足的ResNet教师模型，动态阈值、逐类课程、软置信度权重等一系列选择规则曾回答了这一问题。自监督基础编码器改变了这一格局：在使用DINOv2教师模型时，置信度趋于饱和，因此曾帮助弱教师模型的过滤策略反而可能伤害强教师模型。我们提出CW-BASS v2——一种饱和感知的伪标签选择方法，它读取教师模型的置信度状态，而非固守单一规则。它将留出校准（一种无偏的逐类噪声估计）与一个自适应置信度下限相结合，后者可证明地将保留率约束在远离1的范围内，并将二者组合为单遍门控：在留出切片上度量教师模型置信集合的可靠性（pi_kept = Pr[正确 | c ≥ tau]），当pi_kept ≥ tau满足所要求的置信度时进行严格过滤，否则回退到自适应下限。该边界是预先存在的操作阈值，而非针对mIoU调校的值；在六个DINOv2教师模型上，它能在盲测中正确判断应选择严格过滤还是下限。因此，CW-BASS v2在置信度饱和的基准上通过选择严格过滤恢复了UniMatch V2的操作点（Pascal VOC 1/8上为87.4，对比其报告的87.9；Cityscapes上相差不超过0.5），并在置信集合不可靠的情形下（pi_kept约89%，ADE20K）对其有所改进，此时下限方法略微领先（单一种子，+1.5 mIoU）。这一门控是有原则依据的，因为它所避免的失败是经度量而非假设的：在可靠且饱和的教师模型上，置信度分布的动态范围发生坍缩（Pascal中98%的像素置信度≥0.95），因此自适应截断会淹没保留掩码，自训练也随之退化为确认偏差。

English

Semi-supervised semantic segmentation has long turned on one question, which pseudo-labels to trust, and a generation of selection rules, dynamic thresholds, per-class curricula, soft confidence weights, answered it for the noisy, under-confident ResNet teachers of their day. Self-supervised foundation encoders change the regime: with a DINOv2 teacher, confidence saturates, so the filtering that helped a weak teacher can hurt a strong one. We propose CW-BASS v2, a saturation-aware pseudo-label selection method that reads the teacher's confidence regime rather than committing to one rule. It pairs held-out calibration, an unbiased per-class noise estimate, with a self-adaptive confidence floor that provably bounds retention away from 1, and combines them in a one-pass gate: measure the reliability of the teacher's confident set, pi_kept = Pr[correct | c >= tau], on a held-out slice, and filter strictly when it meets the confidence demanded (pi_kept >= tau), falling back to the adaptive floor otherwise. The boundary is the pre-existing operating threshold, not a value tuned to mIoU, and across six DINOv2 teachers it makes the correct strict-vs-floor call blind. CW-BASS v2 thus recovers the UniMatch V2 operating point on the saturated benchmarks by selecting strict (Pascal VOC 1/8 87.4 against its reported 87.9; Cityscapes within 0.5), and improves on it where the confident set is unreliable (pi_kept ~ 89%, ADE20K), where the floor edges ahead (+1.5 mIoU, single seed). The gate is principled because the failure it avoids is measured, not assumed: on a reliable, saturated teacher the confidence distribution's dynamic range collapses (98% of Pascal pixels >= 0.95), so an adaptive cutoff floods the retention mask and self-training decays into confirmation bias.