Towards Robust Pseudo-Label Learning in Semantic Segmentation: An Encoding Perspective

TL;DR

This work tackles the instability of pseudo-label learning in semantic segmentation under label scarcity by reframing class representation through error-correcting output codes (ECOC). It introduces ECOCSeg, which decouples classes into binary attributes via ECOC, leverages bit-level denoising with reliable-bit mining, and applies customized losses to enforce intra-class compactness and inter-class separation. Theoretical results show ECOC can match one-hot performance in fully supervised settings and offer tighter noise-robustness bounds under appropriate code distance. Empirically, ECOCSeg yields consistent improvements across unsupervised domain adaptation and semi-supervised learning benchmarks, while remaining a flexible, plug-and-play enhancement to existing frameworks.

Abstract

Pseudo-label learning is widely used in semantic segmentation, particularly in label-scarce scenarios such as unsupervised domain adaptation (UDA) and semisupervised learning (SSL). Despite its success, this paradigm can generate erroneous pseudo-labels, which are further amplified during training due to utilization of one-hot encoding. To address this issue, we propose ECOCSeg, a novel perspective for segmentation models that utilizes error-correcting output codes (ECOC) to create a fine-grained encoding for each class. ECOCSeg offers several advantages. First, an ECOC-based classifier is introduced, enabling model to disentangle classes into attributes and handle partial inaccurate bits, improving stability and generalization in pseudo-label learning. Second, a bit-level label denoising mechanism is developed to generate higher-quality pseudo-labels, providing adequate and robust supervision for unlabeled images. ECOCSeg can be easily integrated with existing methods and consistently demonstrates significant improvements on multiple UDA and SSL benchmarks across different segmentation architectures. Code is available at https://github.com/Woof6/ECOCSeg.

Figures (12)

• Figure 1: Comparison of two label encoding methods. (a) Examples of erroneous pseudo labels. (b) Existing methods perform pixel-level classification using argmax-based one-hot encoding. (c) The proposed ECOCSeg predicts the multi-bit binary encoding, which disentangles the classes into fine-grained attributes and enhances the stability of the training process in pseudo-label learning.
• Figure 2: Demonstration of different forms of assigning pseudo labels introduced by ECOCSeg.
• Figure 2: SSL performance on Pascal. The $321$ and $513$ denote the training resolution.
• Figure 3: Pipeline illustration of ECOCSeg. We introduce a new perspective for semantic segmentation (Sec. \ref{['sec:3.3']}), propose a reliable bit mining algorithm to refine the pseudo label (Sec. \ref{['sec:3.4']}), and develop customized optimization criteria (Sec. \ref{['sec:3.5']}).
• Figure 4: (a) Bit count curves under different $T$. (b) Visualization of 2-nearest codewords and confidence map; dotted boxes indicate confusing areas.

Theorems & Definitions (11)

• Theorem 4.1: ECOC Performance in Fully Supervised Setting
• Remark 1
• Theorem 4.2: ECOC Robustness in Pseudo-Label Learning
• Remark 2
• Lemma A.5: NTK convergence, jacot2018neural
• Lemma A.6: Hidden Layer Output Bound, yuerror
• proof : Proof of Theorem \ref{['theorem1']}
• Lemma A.7: Concentration of noisy functions
• Lemma A.8: Binary classifier perturbation bound
• proof