Unsupervised Domain Adaptive Lane Detection via Contextual Contrast and Aggregation

TL;DR

This work tackles unsupervised domain adaptation for lane detection by addressing two bottlenecks: learning discriminative lane feature representations and transferring cross-domain context. It introduces DACCA, which combines a cross-domain contrastive loss with domain-level feature aggregation to fuse pixel-level and domain-level cues, thereby improving cross-domain knowledge transfer. The approach uses two Positive Sample Memory Modules to store lane-domain features and employs a sampling policy to robustly assign positives, while the domain-level feature aggregation module enhances cross-domain context by integrating domain signals into pixel representations. Experiments across six datasets show that DACCA achieves state-of-the-art results in multiple domain-transfer settings, including notably strong performance on CULane→Tusimple and related transfers, demonstrating both effectiveness and generalizability for domain-adaptive lane detection.

Abstract

This paper focuses on two crucial issues in domain-adaptive lane detection, i.e., how to effectively learn discriminative features and transfer knowledge across domains. Existing lane detection methods usually exploit a pixel-wise cross-entropy loss to train detection models. However, the loss ignores the difference in feature representation among lanes, which leads to inefficient feature learning. On the other hand, cross-domain context dependency crucial for transferring knowledge across domains remains unexplored in existing lane detection methods. This paper proposes a method of Domain-Adaptive lane detection via Contextual Contrast and Aggregation (DACCA), consisting of two key components, i.e., cross-domain contrastive loss and domain-level feature aggregation, to realize domain-adaptive lane detection. The former can effectively differentiate feature representations among categories by taking domain-level features as positive samples. The latter fuses the domain-level and pixel-level features to strengthen cross-domain context dependency. Extensive experiments show that DACCA significantly improves the detection model's performance and outperforms existing unsupervised domain adaptive lane detection methods on six datasets, especially achieving the best performance when transferring from CULane to Tusimple (92.10% accuracy), Tusimple to CULane (41.9% F1 score), OpenLane to CULane (43.0% F1 score), and CULane to OpenLane (27.6% F1 score).

Figures (9)

• Figure 1: Diagrams of our DACCA. (a) The main flowchart, and (b) the proposed cross-domain contrastive loss, where DFA is the abbreviation of domain-level feature aggregation. Different colors in the original pixel features represent different lane features.
• Figure 2: An overview of DACCA's framework. (a) Training pipeline of DACCA. (b) Student/Teacher model structure. The source domain-level feature assignment shares the same structure with the target domain-level feature assignment, except that a PSMM saves features from the source domain. The representation head $U$ is used to obtain the pixel-wise feature representation.
• Figure 3: Location of unreliable background pixels in green.
• Figure 4: (a) Comparison among existing contrastive loss variants. (b) Comparison among existing cross-domain context aggregation. (c) Study of the number of anchors. (d) Study of the number of negative samples.
• Figure 5: T-SNE visualization of the key components. (a) SCNN trained with only source domain. (b) SCNN trained with SCCL and TCCL. (c) SCNN trained with SCCL, TCCL, and DFA. Blue and green color represent the source and target domain, respectively.