CODA: A Real-World Road Corner Case Dataset for Object Detection in Autonomous Driving

Li, Kaican; Chen, Kai; Wang, Haoyu; Hong, Liang; Ye, Chaoqiang; Han, Jianhua; Chen, Yukuai; Zhang, Wei; Xu, Chunjing; Yeung, Dit–Yan; Liang, Xiaodan; Li, Zhenguo; Xu, Hang

doi:10.1007/978-3-031-19839-7_24

Cited by 35 publications

(2 citation statements)

References 43 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Zenseact Open Dataset is characterized on large scale and diverse multimodal dataset collected over two years in various European countries larger than 9× that of existing datasets. The CODA [17] dataset addresses limitations in contemporary object detection methods for autonomous driving. It is focusing on the challenge of detecting uncommon objects and corner cases including 30 object categories.…”

Section: Related Workmentioning

confidence: 99%

DataSet for Unflappable Driving : UNFLAPSet

Kim,

Kang,

Yoon

2024

Preprint

View full text Add to dashboard Cite

Autonomous vehicles must be aware of dynamic and static objects, road lanes, road signs and road markings. Recent autonomous vehicles awareness studies adaptable to various road environments continues, it is necessary to construct datasets that accurately reflect the real driving environment. The existing datasets consist of annotations that focus on dynamic and static objects, lanes, and road signs in the driving environment. These annotations enable management of object distance and avoidance, lane recognition and maintenance and awareness of road signs. Although road markings on the road surface provide the traffic regulations and guidance for driving lanes in the driving environment, there is a lack of the road marking datasets containing various types of directions and regulations. When driving without recognizing road markings, difficulty in recognizing lane information poses challenges in determining the appropriate lanes for the driving route and the limitation in predicting the movement of surrounding cars presents difficulties in maintaining stable driving responses. This paper presents a road marking dataset, UNFLAPSet(UNFLAPpable Set). UNFLAPSet is able to be aware of multidimensional information by including the various types of direction and regulations compared to existing road marking datasets. UNFLAPSet consists of three classes - 1. Driving Direction Centric class(DDCclass) 2. Capable of Lane Maneuver class(CLMclass) 3. Specific Condition Caution class(SCCclass), which are based on the meaning of road markings for emphasizing the primary implication of labels unlike the existing road marking datasets. Especially, Merge Line, Merge Arrow and Uturn Dot Line of CLMclass enable the prediction of surrounding vehicles’ movement and subsequently allow for stable responses. Furthermore, restricted direction road markings of SCCclass helps mitigate the risk of accidents caused by driving in unsuitable lanes, thereby facilitating the maintenance of smooth traffic flow. The validation results of UNFLAPSet showed high recognition accuracy for each label, enabling predictable driving integrated perception of driving lanes and surrounding lanes based on road surface marking recognition using UNFLAPSet.

show abstract

Section: Related Workmentioning

confidence: 99%

DataSet for Unflappable Driving : UNFLAPSet

Kim,

Kang,

Yoon

2024

Preprint

View full text Add to dashboard Cite

show abstract

“…In other words, few samples cannot imply the unimportance of the tail classes [6]. Even more, misclassification of tail classes can have severe consequences, especially in critical applications such as medical diagnosis [7] or road monitoring [8]. Therefore, it is important to develop methods that can effectively address the long-tailed distribution of data and improve the recognition performance on tail classes particularly.…”

Section: Introductionmentioning

confidence: 99%

Adjusting Logit in Gaussian Form for Long-Tailed Visual Recognition

Li,

Cheung,

et al. 2024

IEEE Trans. Artif. Intell.

View full text Add to dashboard Cite

It is not uncommon that real-world data are distributed with a long tail. For such data, the learning of deep neural networks becomes challenging because it is hard to classify tail classes correctly. In the literature, several existing methods have addressed this problem by reducing classifier bias, provided that the features obtained with long-tailed data are representative enough. However, we find that training directly on long-tailed data leads to uneven embedding space. That is, the embedding space of head classes severely compresses that of tail classes, which is not conducive to subsequent classifier learning. This paper therefore studies the problem of long-tailed visual recognition from the perspective of feature level. We introduce feature augmentation to balance the embedding distribution. The features of different classes are perturbed with varying amplitudes in Gaussian form. Based on these perturbed features, two novel logit adjustment methods are proposed to improve model performance at a modest computational overhead. Subsequently, the distorted embedding spaces of all classes can be calibrated. In such balanced-distributed embedding spaces, the biased classifier can be eliminated by simply retraining the classifier with classbalanced sampling data. Extensive experiments conducted on benchmark datasets demonstrate the superior performance of the proposed method over the state-of-the-art ones. Source code is available at https://github.com/Keke921/GCLLoss. Impact Statement-Long-tailed visual recognition, a burgeoning field within computer vision, holds profound significance in academic discourse. It fosters advancements in real-world applications by addressing challenges posed by imbalanced datasets, thereby facilitating improved model generalization. In this paper, we propose a simple yet effective logit adjustment method, applicable across different models. Our work provides comprehensive discussions of the proposed method for longtail learning, considering aspects of optimization and geometric interpretation. These discussions contribute to a deeper understanding of long-tail learning and a novel approach for enhancing generalization on the test set. In scholarly pursuits, long-tailed visual recognition underscores the necessity for nuanced and

show abstract