Unet 系列论文一句话总结

Unet 系列论文一句话总结

• Unet 是处理二维数据分割的
• 3D Unet 及 V-Net 是处理三维数据分割的
• 二维网络与三维网络的区别在于，二维网络的卷积核是 2D 的，三维网络的卷积核是 3D 的。所以可以将 3D Unet 看作是 2D Unet 的扩展，只是将 2D 卷积核扩展为 3D 卷积核。

• 疑点：二维注意力机制要怎么扩展到三维？
• 解答：二维注意力机制是在 2D 卷积核的基础上，加入了一个 1x1 的卷积核，将通道数降低，然后再加入一个 1x1 的卷积核，将通道数恢复。所以，可以将 2D 注意力机制看作是 2D 卷积核的扩展，只是将 2D 卷积核扩展为 2D 卷积核。所以，二维注意力机制可以直接扩展到三维。

U-Net: Convolutional Networks for Biomedical Image Segmentation

Ronneberger, O., Fischer, P., & Brox, T. (2015). U-Net: Convolutional Networks for Biomedical Image Segmentation. ArXiv. /abs/1505.04597
https://arxiv.org/pdf/1505.04597v1.pdf
The architecture consists of a contracting path to capture context and a symmetric expanding path that enables precise localiza- tion.

3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation

Çiçek, Ö., Abdulkadir, A., Lienkamp, S. S., Brox, T., & Ronneberger, O. (2016). 3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation. ArXiv. /abs/1606.06650
https://arxiv.org/pdf/1606.06650.pdf
The proposed network extends the previous u-net architecture from Ronneberger et al. by replacing all 2D operations with their 3D counterparts.

V-Net: Fully Convolutional Neural Networks for Volumetric Medical Image Segmentation

Milletari, F., Navab, N., & Ahmadi, S. (2016). V-Net: Fully Convolutional Neural Networks for Volumetric Medical Image Segmentation. ArXiv. /abs/1606.04797
https://arxiv.org/pdf/1606.04797v1.pdf
In this work we propose an approach to 3D image segmentation based on a volumetric, fully convolutional, neural network.

---

引入注意力机制

Attention U-Net: Learning Where to Look for the Pancreas

Oktay, O., & Lee, A. (2018). Attention U-Net: Learning Where to Look for the Pancreas. ArXiv. /abs/1804.03999
https://arxiv.org/pdf/1804.03999.pdf
implementation: https://github.com/LeeJunHyun/Image_Segmentation/
AGs can be easily integrated into standard CNN architectures such as the U-Net model with minimal compu- tational overhead while increasing the model sensitivity and prediction accuracy.

Position-prior Clustering-based Self-attention Module for Knee Cartilage Segmentation

Liang, D., Liu, J., Wang, K., Luo, G., Wang, W., & Li, S. (2022). Position-prior Clustering-based Self-attention Module for Knee Cartilage Segmentation. ArXiv. /abs/2206.10286
https://arxiv.org/pdf/2206.10286
The attention module can capture various scale context depen- dencies and strengthen the relevant feature information to obtain a more ac- curate result in segmentation tasks.