Section VI : INSTANCE SEGMENTATION USING DEEPER U-NETS

[nikhilrayaprolu]

• Figure 9 is too small, please use the same style as Dave Luo's section (Figure 5)
• Reviewers Note (Please clarify these in your section content and confirm):
  How to verify the superiority of the deeper architecture of deeper U-nets over the original architecture of U-net? Does the deeper architecture is necessary for the instance segmentation?

[janickrohrbach]

Recreated figure 9

[janickrohrbach]

Clarification of the superiority of deeper U-Nets compared to the original U-Net (added to the manuscript):

As the work of \cite{li2017deepunet} has impressively shown, U-Nets benefit greatly from a deeper model architecture. It allows the model to make much more detailed segmentations. Particularly near the object borders the predictions of a deeper U-Net tend to be more accurate.

@nikhilrayaprolu can you please add this to the Zotero bib?

@misc{li2017deepunet,
    title={DeepUNet: A Deep Fully Convolutional Network for Pixel-level Sea-Land Segmentation},
    author={Ruirui Li and Wenjie Liu and Lei Yang and Shihao Sun and Wei Hu and Fan Zhang and Wei Li},
    year={2017},
    eprint={1709.00201},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}