Adapt, But Don't Forget: Fine-Tuning and Contrastive Routing for Lane Detection under Distribution Shift

2025-07-22

AI Summary

To address catastrophic forgetting caused by cross-dataset distribution shifts in lane detection, this paper proposes a branched adaptive fine-tuning framework. The method introduces a lightweight parallel branch architecture, integrating component-wise selective fine-tuning with a supervision-guided contrastive learning-driven dynamic routing mechanism to enable distribution-aware assignment of input samples to specialized adaptation branches. By preserving source-domain knowledge stability, the framework significantly enhances generalization on target domains while requiring far fewer parameters than independent multi-model training. Evaluated on multiple heterogeneous target datasets, the approach achieves F1 scores approaching those of full fine-tuning—demonstrating high parameter efficiency, strong transferability, and scalability.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Search and Optimization: Learning to Search
• Computer Vision: Learning & Optimization for CV

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Lane detection models are often evaluated in a closed-world setting, where training and testing occur on the same dataset. We observe that, even within the same domain, cross-dataset distribution shifts can cause severe catastrophic forgetting during fine-tuning. To address this, we first train a base model on a source distribution and then adapt it to each new target distribution by creating separate branches, fine-tuning only selected components while keeping the original source branch fixed. Based on a component-wise analysis, we identify effective fine-tuning strategies for target distributions that enable parameter-efficient adaptation. At inference time, we propose using a supervised contrastive learning model to identify the input distribution and dynamically route it to the corresponding branch. Our framework achieves near-optimal F1-scores while using significantly fewer parameters than training separate models for each distribution.

Problem

• Addressing catastrophic forgetting in lane detection models during fine-tuning
• Developing parameter-efficient adaptation strategies for target distributions
• Dynamic routing of inputs to appropriate branches using contrastive learning

Innovation

• Fine-tuning selected components for adaptation
• Contrastive learning for input distribution identification
• Dynamic routing to corresponding model branches