【论文阅读】Unet、Unet++、Unet3+系列论文网络与研究思路解读

参考文献
[1]Ronneberger, O., Fischer, P., & Brox, T. (2015, October). U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical image computing and computer-assisted intervention (pp. 234-241). Springer, Cham.
[2]Zhou, Z., Siddiquee, M. M. R., Tajbakhsh, N., & Liang, J. (2018). Unet++: A nested u-net architecture for medical image segmentation. In Deep learning in medical image analysis and multimodal learning for clinical decision support (pp. 3-11). Springer, Cham.
[3]Huang, H., Lin, L., Tong, R., Hu, H., Zhang, Q., Iwamoto, Y., … & Wu, J. (2020, May). Unet 3+: A full-scale connected unet for medical image segmentation. In ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) (pp. 1055-1059). IEEE.

文章目录

• 1. 分割中的基本问题
• 2.1 Unet
• 2.2 Unet++
• 2.3 Unet3+
• 3. 小结

1. 分割中的基本问题

分割任务的挑战：

• 看得深 vs 看得细：

  要求网络同时具备抽象与描绘细节的能力
  看得深能准确定位识别物体
  看得细要求能准确区分边缘细节

• 大 vs 小
  要求网络识别多尺度物体的能力
  大物体不至于过分割
  小物体不至于欠分割

2.1 Unet

网络结构设计重点：

• 上采样(encoder)
• 下采样(decoder)
• 跳层连接(Skip connection)

Q1: Is it the feature extractor important？

Q2: Is it the down-sampling important？

Q3: Is it the up-sampling important？

Q4: Is it the skip connection important？

2.2 Unet++

Q1: How many layers is suitable for a CNN？

• 浅层特征：简单特征，如边界，颜色
• 深层特征：抽象特征，如类别

浅有浅的侧重，深有深的优势。

Q2: How to integrate diverse ‘U-net’s with different layers?

存在缺陷：反向求导中间部分未经过，训练不了

Solution：每个尺度都计算损失函数，即deep supervision

Q3: Does the nested and dense skip connection work well? (Ablation study)

Q4: Does the deep supervision work well?(Ablation study)

Q5: Does the deep supervision work well?(剪枝)

2.3 Unet3+

Q1: How to render the Unet-like network utilize multi-scale features?

Ablation study:
Q2: Does the use of multi-scales features work well?
Q3: Does the use of deep supervision work well?

Q4: How to reduce the over-segmentation?

Ablation study:
Q5: Does the CGM work well？

Q6: Why Unet3+ is more efficient with fewer parameters?

3. 小结

本文来自互联网用户投稿，文章观点仅代表作者本人，不代表本站立场，不承担相关法律责任。如若转载，请注明出处。 如若内容造成侵权/违法违规/事实不符，请点击【内容举报】进行投诉反馈！