关于feature的问题 about dcls-sr

[Algolzw]

Hi, a deep neural network (DNN) can be locally linear, but of course it will produce more artifacts than linear network. This problem is mentioned in the DCLS section. To address it , we use the dual path network that inputs both original features and deblurred features.

[Synapsess]

感谢回复，感觉您对自己的paper和project都很上心

[Algolzw]

感谢认可😁

[Synapsess]

据我简单的实验，训练过程中的crash多半是CAlayer引起的。在DANv2中应该是因为两通道之间的乘法操作，而不是通常的cat。一点看法，不一定对。

According my simple experiment,crash is likely caused by CAlayer which is different from DANv2.Unstablilty in DANv2 is caused by multiple operation.