Open meton-robean opened 4 years ago
The critical difference between U-SAID and existing works lies in the high-level component of the cascade. Unlike [29], [12] that placed a pre-trained and fixed segmentation network with true segmentation labels given for training, we design a new unsupervised segmentation awareness (USA) module, that requires no segmentation labels to train with. 这篇作者认为他们没有像下面两个工作(无论是预训练分割网络还是联合训练分割网络),都接触利用语义网络标签信息
12 Zhiwen Fan, Liyan Sun, Xinghao Ding, Yue Huang, Congbo Cai, and
John Paisley. A segmentation-aware deep fusion network for compressed
sensing mri. arXiv preprint arXiv:1804.01210, 2018.
这篇是预训练的语义分割网络帮 医学图像处理任务
截止2019-11-05 该论文是preprint Segmentation-Aware Image Denoising without Knowing True Segmentation
*相当与分割网络可以训练的参数只有最后转成K channels 的那些33卷积层 和上采样层,而且没有用到语义信息标签, 而且特征嵌入网络的参数是固定的,用的是IMAGENET的欲训练,,但是居然可以work!!!**