zp5070 / openlane-v2

[NeurIPS 2023 Track Datasets and Benchmarks] OpenLane-V2: The First Perception and Reasoning Benchmark for Road Driving

OpenLane-V2

The World's First Perception and Reasoning Benchmark for Scene Structure in Autonomous Driving.

• Paper in arXiv (Accepted at NeurIPS 2023 Track Datasets and Benchmarks)
• CVPR 2023 Autonomous Driving Challenge - OpenLane Topology Track
• Point of contact: Huijie (王晖杰) or [email protected]

Introducing OpenLane-V2 Update

We are happy to announce an important update to the OpenLane family, featuring two sets of additional data and annotations.

• Map Element Bucket. We provide a diverse span of road elements (as a bucket) to build the driving scene - on par with all elements in HD Map. Armed with the newly introduced lane segment representations, we unify various map elements to incorporate comprehensive aspects of the captured static scenes to empower DriveAGI.

• Standard-definition (SD) Map. As a new sensor input, SD Map supplements multi-view images with topological and positional priors to strengthen structural acknowledge in the neural networks.

Table of Contents

• News
• Task and Evaluation
• Leaderboard
• Highlights
• Getting Started
• License & Citation
• Related Resources

News

• 2023/08/28 Dataset subset_B released.
• 2023/07/21 Dataset v2.0 and Devkit v2.0.0 released.
• 2023/07/05 The test server of OpenLane Topology is re-opened.
• 2023/06/01 The Challenge at the CVPR 2023 Workshop wraps up.
• 2023/04/21 A baseline based on InternImage released. Check out here.
• 2023/04/20 OpenLane-V2 paper is available on arXiv.
• 2023/02/15 Dataset v1.0, Devkit v1.0.0, and baseline model released.
• 2023/01/15 Initial OpenLane-V2 dataset sample v0.1 released.

(back to top)

Task and Evaluation

Driving Scene Topology

Given sensor inputs, lane segments are required to be perceived, instead of lane centerlines in the task of OpenLane Topology. Besides, pedestrian crossings and road boundaries are also desired to build a comprehensive understanding of the driving scenes. The OpenLane-V2 UniScore (OLUS) is utilized to summarize model performance in all aspects.

OpenLane Topology

Given sensor inputs, participants are required to deliver not only perception results of lanes and traffic elements but also topology relationships among lanes and between lanes and traffic elements simultaneously. In this task, we use OpenLane-V2 Score (OLS) to evaluate model performance.

(back to top)

Leaderboard

Upcoming Challenge in 2024

We plan to release a trailer version of the upcoming challenge. Please stay tuned for more details in Late August.

• Challenge website: AD24Challenge

OpenLane Topology Challenge at CVPR 2023 (Server remains active)

We maintain a leaderboard and test server on the task of scene structure perception and reasoning. If you wish to add new / modify results to the leaderboard, please drop us an email following the instructions here.

• Challenge website: AD23Challenge

(back to top)

Highlights of OpenLane-V2

Unifying Map Representations

One of the superior formulations in the bucket is Lane Segment. It serves as a unifying and versatile representation of lanes, paving the way for multiple downstream applications. With the introduction of SD Map, the autonomous driving system is capable of utilizing these informative priors for achieving satisfactory performance in perception and reasoning.

The following table sums up a detailed comparison of different lane formulations to achieve various functionalities.

Lane Formulation	Functionality
	3D Space	Laneline Cateogry	Lane Direction	Drivable Area	Lane-level Drivable Area	Lane-lane Topology	Bind to Traffic Element	Laneline-less
2D Laneline		✅
3D Laneline	✅	✅
Online (pseudo) HD Map	✅			✅
Lane Centerline	✅		✅			✅	✅	✅
Lane Segment (newly released)	✅	✅	✅	✅	✅	✅	✅	✅

• 3D Space: whether the perceived entities are represented in the 3D space.
• Laneline Category: categories of the visible laneline, such as solid and dash.
• Lane Direction: the driving direction that vehicles need to follow in a particular lane.
• Drivable Area: the entire area where vehicles are allowed to drive.
• Lane-level Drivable Area: drivable area of a single lane, which restricts vehicles from trespassing neighboring lanes.
• Lane-lane Topology: connectivity of lanes that builds the lane network to provide routing information.
• Bind to Traffic Element: correspondence to traffic elements, which provide regulations according to traffic rules.
• Laneline-less: the ability to provide guidance in areas where no visible laneline is available, such as intersections.

Introducing 3D Laneline

Previous datasets annotate lanes on images in the perspective view. Such a type of 2D annotation is insufficient to fulfill real-world requirements. Following the OpenLane-V1 practice, we annotate lanes in 3D space to reflect the geometric properties in the real 3D world.

Recognizing Extremely Small Traffic Elements

Not only preventing collision but also facilitating efficiency is essential. Vehicles follow predefined traffic rules for self-disciplining and cooperating with others to ensure a safe and efficient traffic system. Traffic elements on the roads, such as traffic lights and road signs, provide practical and real-time information.

Topology Reasoning between Lane and Road Elements

A traffic element is only valid for its corresponding lanes. Following the wrong signals would be catastrophic. Also, lanes have their predecessors and successors to build the map. Autonomous vehicles are required to reason about the topology relationships to drive in the right way.

(back to top)

Getting Started

• Download Data
• Install Devkit
• Prepare Dataset
• Train a Model

(back to top)

License & Citation

Prior to using the OpenLane-V2 dataset, you should agree to the terms of use of the nuScenes and Argoverse 2 datasets respectively. OpenLane-V2 is distributed under CC BY-NC-SA 4.0 license. All code within this repository is under Apache License 2.0.

Please use the following citation when referencing OpenLane-V2:

@article{wang2023openlanev2,
  title={OpenLane-V2: A Topology Reasoning Benchmark for Scene Understanding in Autonomous Driving}, 
  author={Wang, Huijie and Li, Tianyu and Li, Yang and Chen, Li and Sima, Chonghao and Liu, Zhenbo and Wang, Yuting and Jiang, Shengyin and Jia, Peijin and Wang, Bangjun and Wen, Feng and Xu, Hang and Luo, Ping and Yan, Junchi and Zhang, Wei and Li, Hongyang},
  journal={arXiv preprint arXiv:2304.10440},
  year={2023}
}

@article{li2023toponet,
  title={Graph-based Topology Reasoning for Driving Scenes},
  author={Li, Tianyu and Chen, Li and Wang, Huijie and Li, Yang and Yang, Jiazhi and Geng, Xiangwei and Jiang, Shengyin and Wang, Yuting and Xu, Hang and Xu, Chunjing and Yan, Junchi and Luo, Ping and Li, Hongyang},
  journal={arXiv preprint arXiv:2304.05277},
  year={2023}
}

(back to top)

Related Resources

• DriveAGI | DriveLM | OpenScene
• TopoNet
• PersFormer | OpenLane
• BEV Perception Survey & Recipe | BEVFormer

(back to top)