Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Control Map Distribution using Map Query Bank for Online Map Generation

Ziming Liu, Leichen Wang, Ge Yang, Xinrun Li, Xingtao Hu, Hao Sun, Guangyu Gao

TL;DR

This paper addresses the high cost of maintaining HD maps by improving visual-based online map generation (OMG) through a learnable Map Query Bank (MQBank) and SD map priors. MQBank decomposes the map distribution into a grid of query representations, with per-scenario initialization generated from augmented SD map data, and introduces a point-level MQBank attention to preserve fine-grained information during BEV feature interaction. It also analyzes SD map data quality and semantic content, proposing a rectification toolchain and demonstrating that SD map semantics improve prediction performance. Empirical results on OpenLaneV2 show state-of-the-art gains in lane and pedestrian-area mAP and topology accuracy, especially with fewer queries, highlighting improved efficiency and robustness for online HD map generation.

Abstract

Reliable autonomous driving systems require high-definition (HD) map that contains detailed map information for planning and navigation. However, pre-build HD map requires a large cost. Visual-based Online Map Generation (OMG) has become an alternative low-cost solution to build a local HD map. Query-based BEV Transformer has been a base model for this task. This model learns HD map predictions from an initial map queries distribution which is obtained by offline optimization on training set. Besides the quality of BEV feature, the performance of this model also highly relies on the capacity of initial map query distribution. However, this distribution is limited because the limited query number. To make map predictions optimal on each test sample, it is essential to generate a suitable initial distribution for each specific scenario. This paper proposes to decompose the whole HD map distribution into a set of point representations, namely map query bank (MQBank). To build specific map query initial distributions of different scenarios, low-cost standard definition map (SD map) data is introduced as a kind of prior knowledge. Moreover, each layer of map decoder network learns instance-level map query features, which will lose detailed information of each point. However, BEV feature map is a point-level dense feature. It is important to keep point-level information in map queries when interacting with BEV feature map. This can also be solved with map query bank method. Final experiments show a new insight on SD map prior and a new record on OpenLaneV2 benchmark with 40.5%, 45.7% mAP on vehicle lane and pedestrian area.

Control Map Distribution using Map Query Bank for Online Map Generation

TL;DR

This paper addresses the high cost of maintaining HD maps by improving visual-based online map generation (OMG) through a learnable Map Query Bank (MQBank) and SD map priors. MQBank decomposes the map distribution into a grid of query representations, with per-scenario initialization generated from augmented SD map data, and introduces a point-level MQBank attention to preserve fine-grained information during BEV feature interaction. It also analyzes SD map data quality and semantic content, proposing a rectification toolchain and demonstrating that SD map semantics improve prediction performance. Empirical results on OpenLaneV2 show state-of-the-art gains in lane and pedestrian-area mAP and topology accuracy, especially with fewer queries, highlighting improved efficiency and robustness for online HD map generation.

Abstract

Reliable autonomous driving systems require high-definition (HD) map that contains detailed map information for planning and navigation. However, pre-build HD map requires a large cost. Visual-based Online Map Generation (OMG) has become an alternative low-cost solution to build a local HD map. Query-based BEV Transformer has been a base model for this task. This model learns HD map predictions from an initial map queries distribution which is obtained by offline optimization on training set. Besides the quality of BEV feature, the performance of this model also highly relies on the capacity of initial map query distribution. However, this distribution is limited because the limited query number. To make map predictions optimal on each test sample, it is essential to generate a suitable initial distribution for each specific scenario. This paper proposes to decompose the whole HD map distribution into a set of point representations, namely map query bank (MQBank). To build specific map query initial distributions of different scenarios, low-cost standard definition map (SD map) data is introduced as a kind of prior knowledge. Moreover, each layer of map decoder network learns instance-level map query features, which will lose detailed information of each point. However, BEV feature map is a point-level dense feature. It is important to keep point-level information in map queries when interacting with BEV feature map. This can also be solved with map query bank method. Final experiments show a new insight on SD map prior and a new record on OpenLaneV2 benchmark with 40.5%, 45.7% mAP on vehicle lane and pedestrian area.

Paper Structure

This paper contains 23 sections, 10 equations, 4 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. BEV network
  4. Online map generation
  5. SD map prior
  6. Proposed method
  7. Review query-based BEV network
  8. Model structure
  9. Optimization
  10. Overall structure of proposed model
  11. Definition of Map query bank (MQBank)
  12. Map query initialization with SD map prior
  13. A new attention module with map query bank
  14. Experiments
  15. Rethink SD map data
  16. ...and 8 more sections

Figures (4)

  • Figure 1: Online map generation model using map query bank.
  • Figure 2: SD map rectification on OpenlaneV2 OpenLaneV2-NeurIPS-2023 dataset. Colorful lines are SD map routes, gray lines are HD map driving centerlines. $\xrightarrow{}$ is to modify the number of lane on this road, which is missing or incorrect. $\times$ is to remove an extra road on SD map. Frame ID is recorded below each map. Please refer to https://github.com/LaoWangBosch/Map_Query_Bank for more details.
  • Figure 3: Comparison of distributions of different map query initialization methods. Each point is a map query feature.
  • Figure 4: Comparison with or without SD map prior. Three metrics (Lane mAP, Pedestrian area mAP, Topology accuracy of lanes) are recorded, using different numbers of queries (50, 100, 200, 400, 800).