《深度学习——匹配》学习参考

机器视觉特征提取：HOG、SIFT、SURF、ORB、LBP、HAAR
(https://mp.weixin.qq.com/s/DGV23WnMsq9o-Fnb0NKs5g)
特征提取、特征描述、特征匹配的通俗解释:
(https://www.pianshen.com/article/8410550640/)

Homography matrix单应性矩阵：
(https://cloud.tencent.com/developer/article/1084332);
(https://zhuanlan.zhihu.com/p/74597564);

SuperPoint特征点提取+特征描述：
SuperPoint:Self-Supervised Interest Point Detection and Description
(https://blog.csdn.net/luo3300612/article/details/88941436)
(https://vincentqin.tech/posts/superpoint/)
CV——图像匹配调研:
(https://blog.csdn.net/Dedication_/article/details/117015487)
UCN:
(https://www.jianshu.com/p/432f527d3961)

SuperGlue特征匹配网络：
(https://vincentqin.tech/posts/superglue/)
(https://zhuanlan.zhihu.com/p/146389956)
(https://blog.csdn.net/weixin_42730997/article/details/108562592)
(https://blog.csdn.net/shizhuoduao/article/details/107120805)

最优算法(OT)讲解:
(https://blog.csdn.net/qq_43173635/article/details/121082676)
OT-Sinkhorm算法：
(https://michielstock.github.io/posts/2017/2017-11-5-OptimalTransport/);
(https://blog.csdn.net/zsfcg/article/details/112510577)

LoFTR：一种基于Transformers的无需特征提取的特征匹配器，提出的LoFTR利用self / cross attention层实现了将局部特征转换为与上下文和位置有关的量，这使得匹配器在稀疏纹理也能得到匹配。
(https://zhuanlan.zhihu.com/p/363037204)
(https://blog.csdn.net/qq_17783559/article/details/120599526)

Linear attention:
(https://jishuin.proginn.com/p/763bfbd2cbeb)