search

junyanz / interactive-deep-colorization

Deep learning software for colorizing black and white images with a few clicks.
https://richzhang.github.io/ideepcolor/
MIT License
2.69k stars 447 forks source link

more details about training the global networks #49

Closed cccusername closed 5 years ago

cccusername commented 6 years ago

hi, i am confused about the input and output when you training the global networks. I will very appreciate that if you can tell me, thank you！

cccusername commented 6 years ago

@richzhang @junyanz

richzhang commented 6 years ago

The input is the global color histogram and the grayscale channel. The output is the color channels.

cccusername commented 6 years ago

Could you share the training code？Thank you @richzhang

richzhang commented 6 years ago

I'm working on a pytorch version right now. I'll probably release the local hints training first. Thanks for your patience.

richzhang commented 6 years ago

Local Hints network training in PyTorch is in this repository. Thanks for your patience.

cccusername commented 6 years ago

thanks a lot @richzhang

Peter2021 commented 5 years ago

@junyanz @richzhang about global hints network hi, i am confused about the loss in global hints network.

I can't find any details about global network's loss in this repository. When I see local hints network in pytorch version, I find huber loss were abandoned in training. The other thing, is about network. I reimplement global hint network in pytorch based on this repository and I use main branch in local hints network in pytorch version. I choose huber loss or L1 loss, but I get yellowish results in visdom. By the way, I use expand_as() function in pytorch to expand global hints. Is necessary tor write its derivation process? I will very appreciate that if you can tell me, thank you！