티스토리 뷰

1. 인용 논문  
Traffic-sign detection and classification in the wild  
Zhe Zhu, Dun Liang, Songhai Zhang, Xiaolei Huang, Baoli Li, Shimin Hu;   
The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 2110-2118  

2. 인용 부분

- Page 1~2

In 2016, Zhu et al. [16] proposed the Tsinghua-Tencent 100K dataset, the largest and most challenging traffic sign dataset. The Tsinghua-Tencent 100K dataset includes 100,000 images in 100 classes and 30,000 traffic sign instances. Compared with the GTSRB and GTSDB datasets, each image has a higher resolution (2,048×2,048), and each traffic sign instance generally occupies a smaller proportion of the image, e.g., 1%, making the Tsinghua-Tencent 100K dataset more suitable for the task of small traffic sign recognition.

 

- Page 3

In the literature [16], a pipeline based on a fully convolutional network was designed to perform both detection and classification simultaneously. Unfortunately, the aforementioned methods do not work well in real-world applications,
primarily because of the inadequacy of GTSDB and GTSRB datasets.


- Page 5

The traffic sign detection performance of the proposed method is evaluated by the standard detection metrics of
recall and precision, which are the same as those used in the previous work [16].

 

To validate the effectiveness of the proposed method, we compare MR-CNN with the original Faster R-CNN [23],
SSD [25], the method proposed by Zhu et al. [16], and the feature pyramid network (FPN) [34].

 

 

- Page 7

댓글
공지사항
최근에 올라온 글
최근에 달린 댓글
Total
Today
Yesterday
링크
«   2024/05   »
1 2 3 4
5 6 7 8 9 10 11
12 13 14 15 16 17 18
19 20 21 22 23 24 25
26 27 28 29 30 31
글 보관함