search

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

Some detail discussion about DIP #57

Open circlehy opened 5 years ago

circlehy commented 5 years ago

Does anybody know why using 32 channels of noise for training, and the width and height of the target image also crop to an integer multiple of 32? Does it mean if we use three channels of noise as input the width and height of the target image need to crop to an integer multiple of 3? We can share more discussion and insight here about DIP.

drrobincroft commented 1 year ago

It seems if any one of the image size is less than 64, then the program would crash.