Learning Lightweight Lane Detection CNNs by Self Attention Distillation

Hou, Yuenan; Ma, Zheng; Liu, Chunxiao; Loy, Chen Change

doi:10.48550/arxiv.1908.00821

Cited by 15 publications

(14 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…CNN architecture has then been adopted to extract advanced features in an end-to-end manner. Most lane detection methods follow pixel-level segmentation-based approach (Pizzati et al 2019;Hou 2019;Mamidala et al 2019;Zou et al 2019). These approaches typically generate segmentation results with an encoder-decoder structure and then post-processing them via curve fitting and clustering.…”

Section: Related Workmentioning

confidence: 99%

Laneformer: Object-aware Row-Column Transformers for Lane Detection

Han¹,

Deng²,

Cai³

et al. 2022

Preprint

View full text Add to dashboard Cite

We present Laneformer, a conceptually simple yet powerful transformer-based architecture tailored for lane detection that is a long-standing research topic for visual perception in autonomous driving. The dominant paradigms rely on purely CNN-based architectures which often fail in incorporating relations of long-range lane points and global contexts induced by surrounding objects (e.g., pedestrians, vehicles). Inspired by recent advances of the transformer encoder-decoder architecture in various vision tasks, we move forwards to design a new end-to-end Laneformer architecture that revolutionizes the conventional transformers into better capturing the shape and semantic characteristics of lanes, with minimal overhead in latency. First, coupling with deformable pixel-wise selfattention in the encoder, Laneformer presents two new row and column self-attention operations to efficiently mine point context along with the lane shapes. Second, motivated by the appearing objects would affect the decision of predicting lane segments, Laneformer further includes the detected object instances as extra inputs of multi-head attention blocks in the encoder and decoder to facilitate the lane point detection by sensing semantic contexts. Specifically, the bounding box locations of objects are added into Key module to provide interaction with each pixel and query while the ROI-aligned features are inserted into Value module. Extensive experiments demonstrate our Laneformer achieves state-of-the-art performances on CULane benchmark, in terms of 77.1% F1 score. We hope our simple and effective Laneformer will serve as a strong baseline for future research in self-attention models for lane detection.

show abstract

Section: Related Workmentioning

confidence: 99%

Laneformer: Object-aware Row-Column Transformers for Lane Detection

Han¹,

Deng²,

Cai³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Inspired by Bisenetv2, RailNet is designed to extract semantic information features of images [20]. The encoder of the RailNet replaces the standard convolution operations by the Depth Wise Convolutions (DWconv) to significantly lower the computational cost [21]. The details of the DWconv reducing the calculation cost are shown in Figure 3.…”

Section: Railnetmentioning

confidence: 99%

Accurate and Lightweight RailNet for Real-Time Rail Line Detection

et al. 2021

View full text Add to dashboard Cite

Railway transportation has always occupied an important position in daily life and social progress. In recent years, computer vision has made promising breakthroughs in intelligent transportation, providing new ideas for detecting rail lines. Yet the majority of rail line detection algorithms use traditional image processing to extract features, and their detection accuracy and instantaneity remain to be improved. This paper goes beyond the aforementioned limitations and proposes a rail line detection algorithm based on deep learning. First, an accurate and lightweight RailNet is designed, which takes full advantage of the powerful advanced semantic information extraction capabilities of deep convolutional neural networks to obtain high-level features of rail lines. The Segmentation Soul (SS) module is creatively added to the RailNet structure, which improves segmentation performance without any additional inference time. The Depth Wise Convolution (DWconv) is introduced in the RailNet to reduce the number of network parameters and eventually ensure real-time detection. Afterward, according to the binary segmentation maps of RailNet output, we propose the rail line fitting algorithm based on sliding window detection and apply the inverse perspective transformation. Thus the polynomial functions and curvature of the rail lines are calculated, and rail lines are identified in the original images. Furthermore, we collect a real-world rail lines dataset, named RAWRail. The proposed algorithm has been fully validated on the RAWRail dataset, running at 74 FPS, and the accuracy reaches 98.6%, which is superior to the current rail line detection algorithms and shows powerful potential in real applications.

show abstract

“…PINet [13] combined point estimation and point instance segmentation, but there were limitations in the presence of the local occlusions or unclear lanes. The lightweight model ENet-SAD [14] has been proposed by applying self attention distillation (SAD) methods to the existing ENet [15]. SCNN [16] and ENet-SAD obtain diverse and rich contextual information to solve these limitations.…”

Section: B Lane Detection Networkmentioning

confidence: 99%

Hybrid tracker based optimal path tracking system for complex road environments for autonomous driving

Seo¹,

Lee²,

Shin³

et al. 2021

Preprint

View full text Add to dashboard Cite

Path tracking system plays a key technology in autonomous driving. The system should be driven accurately along the lane and be careful not to cause any inconvenience to passengers. To address such tasks, this paper proposes hybrid tracker based optimal path tracking system. By applying a deep learning based lane detection algorithm and a designated fast lane fitting algorithm, this paper developed a lane processing algorithm that shows a match rate with actual lanes with minimal computational cost. In addition, three modified path tracking algorithms were designed using the GPS based path or the vision based path. In the driving system, a match rate for the correct ideal path does not necessarily represent driving stability. This paper proposes hybrid tracker based optimal path tracking system by applying the concept of an observer that selects the optimal tracker appropriately in complex road environments. The driving stability has been studied in complex road environments such as straight road with multiple 3way junctions, roundabouts, intersections, and tunnels. Consequently, the proposed system experimentally showed the high performance with consistent driving comfort by maintaining the vehicle within the lanes accurately even in the presence of high complexity of road conditions. Code will be available in https://github.com/DGIST-ARTIV.

show abstract

Learning Lightweight Lane Detection CNNs by Self Attention Distillation

Cited by 15 publications

References 16 publications

Laneformer: Object-aware Row-Column Transformers for Lane Detection

Laneformer: Object-aware Row-Column Transformers for Lane Detection

Accurate and Lightweight RailNet for Real-Time Rail Line Detection

Hybrid tracker based optimal path tracking system for complex road environments for autonomous driving

Contact Info

Product

Resources

About