Heuristic Self-Paced Learning for Domain Adaptive Semantic Segmentation under Adverse Conditions

2026-03-25

AI Summary

This work addresses the class bias issue in unsupervised domain adaptation for semantic segmentation under adverse weather conditions, which arises from static curriculum strategies. To this end, it proposes the first reinforcement learning–based dynamic class scheduler that formulates curriculum learning as a sequential decision-making process. A high-dimensional state encoder captures inter-domain discrepancies, while a class-fair policy gradient objective adaptively adjusts the learning order of semantic classes. Furthermore, the approach integrates mixed supervision signals from both source and target domains to enable self-paced learning with improved class balance. The method achieves state-of-the-art performance on three challenging benchmarks—ACDC, Dark Zurich, and Nighttime Driving—and demonstrates strong generalization from synthetic to real-world adverse-weather scenarios.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Computer Vision: Adversarial Attacks & Robustness
• Planning, Routing, and Scheduling: Learning for Planning and Scheduling

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

The learning order of semantic classes significantly impacts unsupervised domain adaptation for semantic segmentation, especially under adverse weather conditions. Most existing curricula rely on handcrafted heuristics (e.g., fixed uncertainty metrics) and follow a static schedule, which fails to adapt to a model's evolving, high-dimensional training dynamics, leading to category bias. Inspired by Reinforcement Learning, we cast curriculum learning as a sequential decision problem and propose an autonomous class scheduler. This scheduler consists of two components: (i) a high-dimensional state encoder that maps the model's training status into a latent space and distills key features indicative of progress, and (ii) a category-fair policy-gradient objective that ensures balanced improvement across classes. Coupled with mixed source-target supervision, the learned class rankings direct the network's focus to the most informative classes at each stage, enabling more adaptive and dynamic learning. It is worth noting that our method achieves state-of-the-art performance on three widely used benchmarks (e.g., ACDC, Dark Zurich, and Nighttime Driving) and shows generalization ability in synthetic-to-real semantic segmentation.

Problem

• unsupervised domain adaptation
• semantic segmentation
• adverse weather conditions
• curriculum learning
• class bias

Innovation

• Self-Paced Learning
• Domain Adaptation
• Semantic Segmentation
• Reinforcement Learning
• Curriculum Learning