Problem
Research questions and friction points this paper is trying to address.
Pseudo-labeling
Semantic segmentation
Imaging noise
Long-tailed distribution
Class imbalance
When Noise Meets Long-Tail: Feature-Threshold Dual Calibration for Robust Pseudo-Labeling
🤖 AI Summary
This work addresses the coupled challenges of label noise and long-tailed distributions in semi-supervised semantic segmentation, which lead to pseudo-label degradation and insufficient supervision for tail classes. To tackle these issues, we propose FTC-Seg, a framework that achieves joint optimization of the feature space and decision thresholds. Specifically, FTC-Seg introduces Orthogonal Prototype Reconstruction (OPR) for feature purification, effectively suppressing noise interference. Furthermore, it incorporates Adaptive Threshold Calibration (ATC) to dynamically adjust per-class confidence thresholds, thereby breaking the vicious cycle of mutual amplification between noise and class imbalance. By synergistically refining feature representations and classification boundaries, FTC-Seg outperforms existing state-of-the-art methods across four benchmark datasets, demonstrating significant improvements in both overall model robustness and segmentation performance for tail categories.
📝 Abstract
Pseudo-labeling has become a cornerstone of learning from unlabeled data in semantic segmentation. Yet its effectiveness drops sharply in real-world scenarios where strong imaging noise and long-tailed class distributions occur together. We trace this failure to a vicious cycle of pseudo-label degradation. Imaging noise entangles foreground and background features, lowering prediction confidence across all classes, while long-tailed distributions leave tail classes with far fewer training samples and inherently lower confidence. Under fixed high-threshold filtering, these tail-class predictions are systematically filtered out, so they receive no supervision from unlabeled data and thus features keep degrading in subsequent iterations. Critically, noise and long-tail are not independent obstacles but mutually amplifying ones, and addressing either alone is insufficient. To break this cycle, we propose FTC-Seg, a Feature-Threshold dual-Calibration framework built on a standard teacher-student framework. At the feature level, Orthogonal Prototype Reconstruction (OPR) uses a set of learnable orthogonal prototypes to residually purify pixel-wise features, widening the margin between weak foreground targets and noisy backgrounds. At the threshold level, Adaptive Threshold Calibration (ATC) dynamically adjusts class-specific thresholds based on learning difficulty and prediction-distribution bias, rescuing low-confidence pseudo-labels of tail classes from systematic exclusion. Extensive experiments on four public benchmarks spanning three distinct noise modalities show that FTC-Seg achieves strong performance against state-of-the-art methods, with particularly substantial gains on tail classes. Our results establish that jointly calibrating features and thresholds is essential for robust pseudo-labeling under compounded noise and class imbalance.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Pseudo-Labeling
Feature-Threshold Dual Calibration
Orthogonal Prototype Reconstruction
Adaptive Threshold Calibration
Long-Tail Semantic Segmentation
🔎 Similar Papers
No similar papers found.
