车道线检测2022新工作,2D、3D都有
车道线检测是自动驾驶中一项基础而重要的任务,学术和工业界一直投入了大量的工作。小汤也一直对车道线检测任务感兴趣,并在公司开发过相关功能。也分享过一些相关的文章:
相关链接(点击进入):
车道线检测新SOTA CLRNet: Cross Layer Refinement Network for Lane Detec
CVPR2022车道线检测Efficient Lane Detection via Curve Modeling
车道线检测新工作VIL-100: A New Dataset and A Baseline Model for Video In
端到端的多任务感知网络HybridNet,性能优于YOLOP
因为前两天在自动驾驶技术交流群4群里有群友询问近期车道线有哪些新工作,所以小汤整理了一下近期的相关工作,给相关方向的朋友分享一下。也欢迎对车道线检测、车位检测、目标检测、深度估计等相关任务感兴趣的同行、朋友,加入技术交流群,大家一起讨论交流。
Ultra Fast Deep Lane Detection with Hybrid Anchor Driven Ordinal Classification(UFLDv2)
发表(录用):TPAMI 2022
单位:浙江大学
论文:https://arxiv.org/abs/2206.07389
代码:https://github.com/cfzd/Ultra-Fast-Lane-Detection-v2

效果demo:

Rethinking Efficient Lane Detection via Curve Modeling
发表(录用):CVPR 2022
单位:上海交大、华东师大、香港城市大学、商汤
论文:https://arxiv.org/abs/2203.02431
代码:https://github.com/voldemortX/pytorch-auto-drive


论文解读及效果demo:(点击进入)
CVPR2022车道线检测Efficient Lane Detection via Curve Modeling
CLRNet: Cross Layer Refinement Network for Lane Detection
发表(录用):CVPR 2022
单位:飞布科技(Fabu)、浙江大学
论文:https://arxiv.org/pdf/2203.10350.pdf
代码:https://github.com/Turoad/CLRNet


论文解读及效果demo:(点击进入)
车道线检测新SOTA CLRNet: Cross Layer Refinement Network for Lane Detec
A Keypoint-based Global Association Network for Lane Detection
发表(录用):CVPR 2022
单位:北大、中科大、商汤
论文:https://arxiv.org/pdf/2204.07335.pdf
代码:https://github.com/Wolfwjs/GANet
提出了一个全局关联网络(GANet)来从一个新的角度描述车道检测问题,其中每个关键点直接回归到车道线的起点,而不是逐点扩展。
具体地说,关键点与其所属车道线的关联是通过预测它们在全局范围内与相应车道起点的偏移量来实现的,而不相互依赖,这可以并行进行,从而大大提高效率。此外,还提出了一种车道感知特征聚合器(LFA),它自适应地捕获相邻关键点之间的局部相关性,以补充全局关联的局部信息。在高FPS的同时,在CULane上的F1分数为79.63%,Tusimple数据集上的F1分数为97.71%。

效果demo:

Eigenlanes: Data-Driven Lane Descriptors for Structurally Diverse Lanes github SDLane Dataset
发表(录用):CVPR 2022
单位:Korea University, dot.ai
论文:https://arxiv.org/abs/2203.15302
代码:https://github.com/dongkwonjin/Eigenlanes

效果demo:

Towards Driving-Oriented Metric for Lane Detection Models
发表(录用):CVPR 2022
单位:University of California
论文:https://arxiv.org/abs/2203.16851
代码:https://github.com/ASGuard-UCI/ld-metric
design 2 new driving-oriented metrics for lane detection: End-to-End Lateral Deviation metric (E2E-LD) is directly formulated based on the requirements of autonomous driving, a core downstream task of lane detection; Per-frame Simulated Lateral Deviation metric (PSLD) is a lightweight surrogate metric of E2E-LD.

ONCE-3DLanes: Building Monocular 3D Lane Detection Homepage github Dataset
发表(录用):CVPR 2022
单位:复旦大学、华为诺亚实验室
论文:https://arxiv.org/pdf/2205.00301.pdf
代码:https://github.com/once-3dlanes/once_3dlanes_benchmark
the largest real-world lane detection dataset published up to now, containing more complex road scenarios with various weather conditions, different lighting conditions as well as a variety of geographical locations.(最大一个)

效果demo:

PersFormer: 3D Lane Detection via Perspective Transformer and the OpenLane Benchmark OpenLane Dataset
发表(录用):暂时未知
单位:Shanghai AI Laboratory, Purdue University, Carnegie Mellon University, SenseTime Research, Shanghai Jiao Tong University
论文:https://arxiv.org/abs/2203.11089
代码:https://github.com/OpenPerceptionX/PersFormer_3DLane
the first real-world, large-scale 3D lane dataset and corresponding benchmark, OpenLane, to support research into the problem.(第一个)

效果demo:
HybridNets: End-to-End Perception Network
发表(录用):暂时未知
单位:Shanghai AI Laboratory, Purdue University, Carnegie Mellon University, SenseTime Research, Shanghai Jiao Tong University
论文:https://arxiv.org/ftp/arxiv/papers/2203/2203.09035.pdf
代码:https://github.com/datvuthanh/HybridNets
效果demo: