Domain-Incremental Semantic Segmentation for Autonomous Driving under Adverse Driving Conditions

2025-01-09

AI Summary

Autonomous driving semantic segmentation suffers from severe performance degradation under adverse conditions (e.g., rain, fog, low illumination) and catastrophic forgetting during domain-incremental learning. To address these dual challenges, this paper proposes a task-agnostic Progressive Semantic Segmentation (PSS) framework. PSS dynamically expands domain-specific segmentation modules and employs a lightweight ensemble of convolutional autoencoders for unsupervised domain identification and precise module routing—eliminating reliance on explicit domain labels. Furthermore, it introduces a multi-granularity domain adaptation evaluation mechanism to jointly preserve source-domain knowledge and enhance target-domain generalization. Evaluated on multiple adverse-condition datasets, PSS achieves significant cross-domain segmentation accuracy gains while maintaining source-domain stability and demonstrating strong generalization to both similar and unseen domains. The framework provides a scalable, modular solution for robust autonomous driving perception.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Application Domains: Mobility, Driving & Flight

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Semantic segmentation for autonomous driving is an even more challenging task when faced with adverse driving conditions. Standard models trained on data recorded under ideal conditions show a deteriorated performance in unfavorable weather or illumination conditions. Fine-tuning on the new task or condition would lead to overwriting the previously learned information resulting in catastrophic forgetting. Adapting to the new conditions through traditional domain adaption methods improves the performance on the target domain at the expense of the source domain. Addressing these issues, we propose an architecture-based domain-incremental learning approach called Progressive Semantic Segmentation (PSS). PSS is a task-agnostic, dynamically growing collection of domain-specific segmentation models. The task of inferring the domain and subsequently selecting the appropriate module for segmentation is carried out using a collection of convolutional autoencoders. We extensively evaluate our proposed approach using several datasets at varying levels of granularity in the categorization of adverse driving conditions. Furthermore, we demonstrate the generalization of the proposed approach to similar and unseen domains.

Problem

• Adverse Driving Conditions
• Autonomous Vehicles
• Environmental Adaptation

Innovation

• Progressive Semantic Segmentation
• Convolutional Autoencoder
• Adaptive Performance Enhancement