添加L1正则来约束BN层系数
语义边缘检测和语义分割的关系调研结果为，语义信息可以用来增强语义分割的效果，也有一定的优点和采用理由，但此类论文的数量并不是很多，语义分割的多数方法还是使用深度学习直接做像素分类。在对比两者的性能时发现，两者的performance指标不一样，所以无法进行对比。
Research findings on the relationship between semantic edge detection and semantic segmentation indicate that semantic information can be used to enhance the effectiveness of semantic segmentation. However, the number of such papers is not very large, and most methods of semantic segmentation still use deep learning for direct pixel classification. When comparing the performance of the two, it is found that their performance metrics are different, making it impossible to compare them directly.
边缘检测和显著性检测的关系是，边缘信息可以用来增强显著性检测的效果。因为基本边缘检测在数据处理时都是第一个步骤。The relationship between edge detection and saliency detection is that edge information can enhance the effect of saliency detection. This is because basic edge detection is typically the first step in data processing.
BSDS500 轮廓检测与语义分割数据集_数据集-阿里云天池
Learning Spatial Context: Using Stuff to Find Things (stanford.edu)
阅读了三篇关于基于深度学习的边缘检测的综述论文，发现了12篇关于更流行的算法的论文。整理了模型架构和指标信息。
After reading three review papers on deep learning-based edge detection, I discovered 12 papers on more popular algorithms. I have organized the model structures and metric information.