Search papers, labs, and topics across Lattice.
This paper introduces CW-BASS v2, a novel pseudo-label selection method for semi-supervised semantic segmentation that adapts to the confidence saturation of self-supervised foundation model teachers like DINOv2. By employing a combination of held-out calibration and a self-adaptive confidence floor, the method effectively distinguishes between reliable and unreliable pseudo-labels, thereby avoiding the pitfalls of traditional filtering techniques. The results demonstrate that CW-BASS v2 not only matches the performance of existing methods on saturated benchmarks but also improves mIoU in scenarios where the confident set is less reliable, achieving notable gains on datasets like ADE20K.
Trusting the wrong pseudo-labels can lead to confirmation bias, but CW-BASS v2 intelligently adapts to the confidence landscape of strong foundation models, yielding significant performance improvements.
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.
