About Weight decay for the network

DmitryUlyanov / deep-image-prior

Image restoration with neural networks but without learning.

https://dmitryulyanov.github.io/deep_image_prior

Other

7.79k stars 1.42k forks source link

About Weight decay for the network #55

Open Shaosifan opened 5 years ago

Shaosifan commented 5 years ago

I find that the weight decay of network is not used in the optimization. So I try to set weight decay (0.0001) in the Adam function, and the PSNR of the reconstruction relative to reference truth begins to fluctuate.

Do you observe this phenomenon? Is it better not to use weight decay in such unsupervised network?

Shaosifan commented 5 years ago

PSNR