Yolov8 actual combat: traffic signal roadsign recognition, by adding CVPR203 DCNV3 and BiLevelRoutingAttention, violent increase

1. Roadsign dataset introduction

Dataset size: 877 images

Category: speedlimit, crosswalk, trafficlight, stop

 2. Roadsign recognition based on YOLOV8

2.1 Original yolov8 performance analysis

The original map is 0.841

 

 

2.1 Join DCNV3

blog address:

https://cv2023.blog.csdn.net/article/details/130224460

map improved from 0.841 to 0.881