Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
156 tokens/sec
GPT-4o
7 tokens/sec
Gemini 2.5 Pro Pro
45 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Lane Graph as Path: Continuity-preserving Path-wise Modeling for Online Lane Graph Construction (2303.08815v3)

Published 15 Mar 2023 in cs.CV and cs.RO

Abstract: Online lane graph construction is a promising but challenging task in autonomous driving. Previous methods usually model the lane graph at the pixel or piece level, and recover the lane graph by pixel-wise or piece-wise connection, which breaks down the continuity of the lane and results in suboptimal performance. Human drivers focus on and drive along the continuous and complete paths instead of considering lane pieces. Autonomous vehicles also require path-specific guidance from lane graph for trajectory planning. We argue that the path, which indicates the traffic flow, is the primitive of the lane graph. Motivated by this, we propose to model the lane graph in a novel path-wise manner, which well preserves the continuity of the lane and encodes traffic information for planning. We present a path-based online lane graph construction method, termed LaneGAP, which end-to-end learns the path and recovers the lane graph via a Path2Graph algorithm. We qualitatively and quantitatively demonstrate the superior accuracy and efficiency of LaneGAP over conventional pixel-based and piece-based methods on the challenging nuScenes and Argoverse2 datasets under controllable and fair conditions. Compared to the recent state-of-the-art piece-wise method TopoNet on the OpenLane-V2 dataset, LaneGAP still outperforms by 1.6 mIoU, further validating the effectiveness of path-wise modeling. Abundant visualizations in the supplementary material show LaneGAP can cope with diverse traffic conditions. Code is released at \url{https://github.com/hustvl/LaneGAP}.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (47)
  1. Roadtracer: Automatic extraction of road networks from aerial images. In CVPR, 2018.
  2. Improved road connectivity by joint learning of orientation and segmentation. In CVPR, 2019.
  3. Learning and aggregating lane graphs for urban automated driving. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13415–13424, 2023.
  4. Fully convolutional network for automatic road extraction from satellite imagery. In CVPR, 2018.
  5. nuscenes: A multimodal dataset for autonomous driving. In CVPR, 2020.
  6. Structured bird’s-eye-view traffic scene understanding from onboard images. In ICCV, 2021.
  7. Topology preserving local road network estimation from single onboard camera image. In CVPR, 2022.
  8. End-to-end object detection with transformers. In ECCV, 2020.
  9. Persformer: 3d lane detection via perspective transformer and the openlane benchmark. In ECCV, 2022.
  10. Efficient and robust 2d-to-bev representation learning via geometry-guided kernel transformer. arXiv preprint arXiv:2206.04584, 2022.
  11. Neural turtle graphics for modeling city road layouts. In ICCV, 2019.
  12. Pivotnet: Vectorized pivot learning for end-to-end hd map construction. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3672–3682, 2023.
  13. You only look at one sequence: Rethinking transformer in vision through object detection. NeurIPS, 2021.
  14. Rethinking efficient lane detection via curve modeling. In CVPR, 2022.
  15. 3d-lanenet: end-to-end 3d multiple lane detection. In ICCV, 2019.
  16. Gen-lanenet: A generalized and scalable approach for 3d lane detection. In ECCV, 2020.
  17. Deep residual learning for image recognition. In CVPR, 2016.
  18. Lane-level street map extraction from aerial imagery. In WACV, 2022.
  19. Sat2graph: Road graph extraction through graph-tensor encoding. In ECCV, 2020.
  20. Hdmapnet: An online hd map construction and evaluation framework. In ICRA, 2022.
  21. Bevformer: Learning bird’s-eye-view representation from multi-camera images via spatiotemporal transformers. In ECCV, 2022.
  22. Topological map extraction from overhead images. In ICCV, 2019.
  23. MapTR: Structured modeling and learning for online vectorized HD map construction. In ICLR, 2023.
  24. Focal loss for dense object detection. In ICCV, 2017.
  25. Learning to predict 3d lane shape and camera pose from a single image via geometry constraints. In AAAI, 2022.
  26. End-to-end lane shape prediction with transformers. In WACV, 2021.
  27. Vectormapnet: End-to-end vectorized hd map learning. arXiv preprint arXiv:2206.08920, 2022.
  28. Petrv2: A unified framework for 3d perception from multi-camera images. arXiv preprint arXiv:2206.01256, 2022.
  29. Vision-based uneven bev representation learning with polar rasterization and surface estimation. arXiv preprint arXiv:2207.01878, 2022.
  30. Bevfusion: Multi-task multi-sensor fusion with unified bird’s-eye view representation. arXiv preprint arXiv:2205.13542, 2022.
  31. Learning ego 3d representation as ray tracing. In ECCV, 2022.
  32. Vision-centric bev perception: A survey. arXiv preprint arXiv:2208.02797, 2022.
  33. Deeproadmapper: Extracting road topology from aerial images. In ICCV, 2017.
  34. Hdmapgen: A hierarchical graph generative model of high definition maps. In CVPR, 2021.
  35. End-to-end vectorized hd-map construction with piecewise bezier curve. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13218–13228, 2023.
  36. U-net: Convolutional networks for biomedical image segmentation. In MICCAI, 2015.
  37. Instagram: Instance-level graph modeling for vectorized hd map learning. arXiv preprint arXiv:2301.04470, 2023.
  38. Keep your eyes on the lane: Real-time attention-guided lane detection. In CVPR, 2021.
  39. Vecroad: Point-based iterative graph exploration for road graphs extraction. In CVPR, 2020.
  40. A keypoint-based global association network for lane detection. In CVPR, 2022.
  41. Argoverse 2: Next generation datasets for self-driving perception and forecasting. arXiv preprint arXiv:2301.00493, 2023.
  42. Centerlinedet: Road lane centerline graph detection with vehicle-mounted sensors by transformer for high-definition map creation. arXiv preprint arXiv:2209.07734, 2022.
  43. icurb: Imitation learning-based detection of road curbs using aerial images for autonomous driving. IEEE Robotics and Automation Letters, 6(2):1097–1104, 2021.
  44. Second: Sparsely embedded convolutional detection. Sensors, 18(10):3337, 2018.
  45. Cross-view transformers for real-time map-view semantic segmentation. In CVPR, 2022.
  46. D-linknet: Linknet with pretrained encoder and dilated convolution for high resolution satellite imagery road extraction. In CVPRW, 2018.
  47. Deformable DETR: deformable transformers for end-to-end object detection. In ICLR, 2021.
Citations (28)
View on Semantic Scholar

Summary

  • The paper presents essential submission requirements, including language standards, page limits, and anonymization protocols.
  • It emphasizes strict adherence to formatting rules and blind review processes to maintain consistency in research evaluation.
  • The guidelines streamline the review process by reducing administrative burdens and ensuring each manuscript meets uniform criteria.

Insights into Submission Guidelines for ICCV Proceedings

The document presented offers a comprehensive guide on formatting and submitting manuscripts for the ICCV (International Conference on Computer Vision) proceedings. The primary focus of the paper is to provide detailed instructions on various aspects of manuscript preparation, ensuring that authors adhere to the specified standards for submission. This overview discusses the notable features and recommendations encapsulated within the guidelines.

Key Aspects of Manuscript Preparation

The document covers the fundamental requirements for language, paper length, formatting, anonymization, and other essential details crucial for the preparation of a manuscript for ICCV.

  1. Language: English is mandated as the language for all manuscripts, reinforcing the uniformity in communication among a global audience of researchers.
  2. Paper Length: Each submission must not exceed eight pages, exclusive of the references. This constraint ensures focused and concise presentation of research findings without extraneous information. Notably, there is no provision for overlength papers, highlighting the importance of adhering to the prescribed limits.
  3. Formatting: The document is meticulous in outlining the required formatting style, including the use of Times Roman for textual elements, spacing, paragraph indentation, and alignment, all configured to optimize readability and professionalism. A consistent two-column format is emphasized for the body text, maintaining uniformity across submissions.
  4. Blind Review Process: The paper clarifies misconceptions regarding blind reviews, explaining that the anonymity applies to referencing work. It advises against wording that could reveal author identities while citing past work, and encourages proper citation to ensure the reviewing committee can fully assess the manuscript’s contributions within the context of previous literature.
  5. Figures and Tables: Recommendations are given on the inclusion and formatting of figures and tables to ensure they are legible and integrate seamlessly with the text. This is crucial for visual clarity and for reviewers or readers who may print the document for better accessibility.
  6. Margins and Page Numbering: Precise margin specifications and page numbering conventions are laid out to maintain consistency and facilitate an efficient review process. While page numbers should be present during reviews, they are omitted in the final camera-ready submissions.

Implications and Future Considerations

These guidelines play a critical role in standardizing submissions, fostering an equitable review environment where all research is presented on equal technical grounds. Such standardization allows reviewers to focus more on the content's merit rather than being distracted by varied presentation styles.

The adherence to explicitly defined instructions minimizes the administrative burden involved in the initial review phase, allowing for a focus on the technical aspects of the submissions. Any deviations tend to result in outright rejection, emphasizing the importance of conforming to these guidelines for those submitting to ICCV.

In terms of future directions, the evolving nature of publication standards might incorporate more automated and user-friendly systems for checking compliance with these stringent guidelines prior to submission. This could further streamline the submission process, reducing errors arising from manual formatting.

The detailed structuring of this document reflects a dynamic understanding of modern publication needs, catering to a diverse and expansive community of computer vision researchers. As ICCV continues to be a pivotal venue for cutting-edge research dissemination, adherence to these guidelines remains essential for contributors seeking to influence and contribute to advancing the field.

File Document Download Save Streamline Icon: https://streamlinehq.com PDF File Document Download Save Streamline Icon: https://streamlinehq.com Markdown
X Twitter Logo Streamline Icon: https://streamlinehq.com

Tweets