DiffusionLane: Diffusion Model for Lane Detection

TL;DR

DiffusionLane treats lane detection as a denoising diffusion process over lane parameters, introducing random lane anchors and a diffusion-based refinement in the parameter space. A hybrid diffusion decoder plus an auxiliary head address encoder representation challenges caused by noisy anchors, enabling robust cross-domain generalization. Empirically, DiffusionLane achieves state-of-the-art or competitive results across four benchmarks (CULane, LLAMAS, Tusimple, Carlane), including 81.32% F1 on CULane, 97.59% F1 on LLAMAS, and strong domain-adaptation performance on MuLane, demonstrating the viability of diffusion-based lane detection. The work highlights diffusion models as a practical baseline for anchor-based lane detection, with potential implications for robust perception in distribution-shift scenarios.

Abstract

In this paper, we present a novel diffusion-based model for lane detection, called DiffusionLane, which treats the lane detection task as a denoising diffusion process in the parameter space of the lane. Firstly, we add the Gaussian noise to the parameters (the starting point and the angle) of ground truth lanes to obtain noisy lane anchors, and the model learns to refine the noisy lane anchors in a progressive way to obtain the target lanes. Secondly, we propose a hybrid decoding strategy to address the poor feature representation of the encoder, resulting from the noisy lane anchors. Specifically, we design a hybrid diffusion decoder to combine global-level and local-level decoders for high-quality lane anchors. Then, to improve the feature representation of the encoder, we employ an auxiliary head in the training stage to adopt the learnable lane anchors for enriching the supervision on the encoder. Experimental results on four benchmarks, Carlane, Tusimple, CULane, and LLAMAS, show that DiffusionLane possesses a strong generalization ability and promising detection performance compared to the previous state-of-the-art methods. For example, DiffusionLane with ResNet18 surpasses the existing methods by at least 1\% accuracy on the domain adaptation dataset Carlane. Besides, DiffusionLane with MobileNetV4 gets 81.32\% F1 score on CULane, 96.89\% accuracy on Tusimple with ResNet34, and 97.59\% F1 score on LLAMAS with ResNet101. Code will be available at https://github.com/zkyntu/UnLanedet.

Figures (6)

• Figure 1: Diffusion model for lane detection. (a) The diffusion model where $q$ denotes the diffusion process and $p_{\theta}$ represents the reverse process. (b) Noise-to-image process from noisy pixels to target image. (c) Our noise-to-lane paradigm from noisy lane anchors to target lanes.
• Figure 2: Illustration of lane representation. We take 7 sampling points as the example.
• Figure 3: Training pipeline of DiffusionLane. The image encoder and SAFPN xiao2023adnet extract the multi-scale image features $M_i,i \in [0,2]$ and feed them into the hybrid diffusion decoder. The hybrid diffusion decoder, containing a stack of hybrid diffusion blocks, refines the noisy lanes to target lanes iteratively. Auxiliary head, only used in the training process, improves the feature representation of the encoder via learnable lane anchors.
• Figure 4: Visualization of the attention map of the encoder.
• Figure 5: Architecture of a hybrid diffusion block.