SparseMeXT Unlocking the Potential of Sparse Representations for HD Map Construction

TL;DR

This paper tackles online HD map construction with sparse representations, addressing inefficiencies that plagued prior sparse methods and the performance gap to dense BEV approaches. It introduces SparseMeXt, a sparse encoder–decoder framework featuring a memory-assisted temporal model, a sparse-to-dense auxiliary segmentation pathway, and a physically guided denoising module (PPDN) that employs rotation, location, scale, and curvature noise aligned with geometric priors. Through architectural optimizations (SiMo neck, tailored backbone pretraining), a decoupled decoder, and a sparse–dense supervision scheme, SparseMeXt achieves state-of-the-art results on NuScenes across Tiny/Base/Large backbones, including 68.9% mAP for Large at over 20 fps, and improves centerline prediction. The approach demonstrates that sparse map representations can outperform dense methods while maintaining real-time performance, signaling a shift in efficiency–accuracy trade-offs for online HD map construction in autonomous driving.

Abstract

Recent advancements in high-definition \emph{HD} map construction have demonstrated the effectiveness of dense representations, which heavily rely on computationally intensive bird's-eye view \emph{BEV} features. While sparse representations offer a more efficient alternative by avoiding dense BEV processing, existing methods often lag behind due to the lack of tailored designs. These limitations have hindered the competitiveness of sparse representations in online HD map construction. In this work, we systematically revisit and enhance sparse representation techniques, identifying key architectural and algorithmic improvements that bridge the gap with--and ultimately surpass--dense approaches. We introduce a dedicated network architecture optimized for sparse map feature extraction, a sparse-dense segmentation auxiliary task to better leverage geometric and semantic cues, and a denoising module guided by physical priors to refine predictions. Through these enhancements, our method achieves state-of-the-art performance on the nuScenes dataset, significantly advancing HD map construction and centerline detection. Specifically, SparseMeXt-Tiny reaches a mean average precision \emph{mAP} of 55.5% at 32 frames per second \emph{fps}, while SparseMeXt-Base attains 65.2% mAP. Scaling the backbone and decoder further, SparseMeXt-Large achieves an mAP of 68.9% at over 20 fps, establishing a new benchmark for sparse representations in HD map construction. These results underscore the untapped potential of sparse methods, challenging the conventional reliance on dense representations and redefining efficiency-performance trade-offs in the field.

Figures (4)

• Figure 2: Overall architecture of SparseMeXt.
• Figure 3: Architecture of the decouple decoder. We independently split decoder to two branch.
• Figure 4: Architecture of the sparse-dense segmentation head.
• Figure 5: Qualitative results of SparseDrive, SparseMeXt, SparseMext LongRange and SparseMeXt Centerline.