简介 Intro
YOLO v4 中国交通标志识别模型是一种专为中国复杂交通环境设计的深度学习实时目标检测系统, 它在 YOLO 算法的基础上进行了优化, 以适应中国特有的交通标志识别需求。该模型通过高效的特征提取网络和多尺度预测机制, 结合注意力机制和改进的空间金字塔池化技术, 显著提升了对不同尺寸和角度交通标志的识别精度, 即使在光照变化和恶劣天气条件下也能保持高准确率。在 CCTSDB 2021 数据集上的测试显示, 模型达到了 96.62% 的检测准确率和 79.73% 的召回率, F-1 分数为 87.37%, mAP 高达 92.77%, 同时保持约 81 帧每秒的高帧率, 满足了智能车辆实时性的需求。该模型广泛应用于智能交通系统、自动驾驶车辆环境感知、城市监控和工业自动化等领域, 为提升道路安全和交通效率提供了技术支撑。
The YOLO v4 Chinese Traffic Sign Recognition model is a deep learning real-time object detection system specifically designed for the complex traffic environment in China, optimized based on the YOLO algorithm to meet the unique needs of traffic sign recognition in China. This model significantly enhances the recognition accuracy of traffic signs of various sizes and angles through an efficient feature extraction network and multi-scale prediction mechanisms, combined with attention mechanisms and improved spatial pyramid pooling techniques, maintaining high accuracy even under varying lighting and adverse weather conditions. Testing on the CCTSDB 2021 dataset showed that the model achieved a detection accuracy of 96.62%, a recall rate of 79.73%, an F-1 score of 87.37%, and an mAP of up to 92.77%, while maintaining a high frame rate of approximately 81 frames per second, meeting the real-time requirements of intelligent vehicles. The model is widely applied in fields such as intelligent transportation systems, environmental perception of autonomous vehicles, urban surveillance, and industrial automation, providing technical support for enhancing road safety and traffic efficiency.
快速使用 Usage
from modelscope import snapshot_download
model_dir = snapshot_download('Genius-Society/yolov4_tt100k')
维护 Maintenance
git clone git@hf.co:Genius-Society/yolov4_tt100k
cd yolov4_tt100k
镜像 Mirror
https://www.modelscope.cn/models/Genius-Society/yolov4_tt100k