Image Registration: From SIFT to Deep Learning - 知乎
不想太消极,但是Dosovitskiy的方法在实践中并没有真正起作用。它甚至没有经过适当的测试。
DeepHomography也是如此。
可以在实践中使用的深度检测器和描述符是:HardNet,SuperPoint和D2Net:请参阅最新和相关的CVPR研讨会
图像匹配:局部功能及其他
关于图像采样的内容:ASIFT和MODS。
关于医学图像注册:ANHIR-大挑战很好的解释,但我将添加到Dmytro。
实际上,配准是在两个图像之间进行的,在这些图像中照明/噪声可能会发生变化。但是生成DeepHomography训练数据的过程没有考虑到这一点。
我建议检查提及的文件:Affnet和HardNet。您没有讨论光流技术,这是一种出色的注册工具。例如,flownet。
Unsupervised Deep Homography: A Fast and Robust Homography Estimation Model
https://arxiv.org/pdf/1709.03966.pdf
Deep Image Homography Estimation
https://blog.csdn.net/zcg1942/article/details/85254077
https://blog.csdn.net/ajing2014/article/details/53998866
https://arxiv.org/abs/1606.03798
https://github.com/yishiliuhuasheng/deep_image_homography_estimation
https://github.com/search?q=Deep+Image+Homography+Estimation
https://hal.archives-ouvertes.fr/hal-01859616/file/cnngeometric_pami.pdf
Convolutional neural network architecture for geometric matching
https://www.youtube.com/watch?v=4RMwGcT4uA8
Spatial Transformer Networks
https://blog.csdn.net/shaoxiaohu1/article/details/51809605
https://www.cnblogs.com/liaohuiqiang/p/9226335.html
https://blog.csdn.net/WZZ18191171661/article/details/100621925
https://zhuanlan.zhihu.com/p/43054073
Joint Multiple Image Parametric Transformation Estimation Via Convolutional Neural Networks
https://ieeexplore.ieee.org/document/8300522?denied=
Repeatability Is Not Enough: Learning Affine Regions via Discriminability
http://cmp.felk.cvut.cz/~radenfil/publications/Mishkin-ECCV18.pdf
https://github.com/ducha-aiki/affnet
谷歌学术搜索引用:
Convolutional neural network architecture for geometric matching
End-to-end weakly-supervised semantic alignment
SuperPoint: Self-Supervised Interest Point Detection and Description
Neighbourhood Consensus Networks
http://papers.nips.cc/paper/7437-neighbourhood-consensus-networks.pdf
http://papers.nips.cc/paper/7437-neighbourhood-consensus-networks
Deep Learning of Graph Matching
http://openaccess.thecvf.com/content_cvpr_2018/papers/Zanfir_Deep_Learning_of_CVPR_2018_paper.pdf
Deep Fundamental Matrix Estimation
Self-Supervised Learning of Geometrically Stable Features Through Probabilistic Introspection
https://arxiv.org/pdf/1905.03561.pdf
D2-Net: A Trainable CNN for Joint Description and Detection of Local Features
https://github.com/AaltoVision/DGC-Net
Geometric Image Correspondence Verification by Dense Pixel Matching