search

yu4u / noise2noise

An unofficial and partial Keras implementation of "Noise2Noise: Learning Image Restoration without Clean Data"
MIT License
1.07k stars 237 forks source link

Computing Resources #23

Open rogershijin opened 5 years ago

rogershijin commented 5 years ago

Could you give an idea of what computing resources you used to train your models and how long training took (along with how many epochs you trained for)?

yu4u commented 5 years ago

I'm using GTX 1080 / 1080Ti. One epoch requires a few minutes. Default training script runs for 60 epochs.

AllenJac commented 5 years ago

I have some questions about the train code, "train.py". When I run the code, all parameters are default, there is no mistake, but the code never stop. So I just use one image to train, and set "epoch=1", it still doesn't stop. Finally, I remove all my training images and set ("--image_dir", default=" ",), there is no path to train, it still can run without any mistakes. Do you know what's the reason and why it doesn't stop even train one image?

yu4u commented 5 years ago

I think I fixed this problem. https://github.com/yu4u/noise2noise/pull/24/ Could you try again after pulling the repository to adopt the PR?

AllenJac commented 5 years ago

Thanks for fixing it. But there is still something weird. I just use one image ( .jpg) from your training dataset to train, and set "epoch=2", other parameters are default, it still doesn't stop (Just keep "Epoch 1/2" and never stop), are there any thing I ignore?

yu4u commented 5 years ago

What is the training image size? If the image size is too small (< 64), training falls into an infinite loop...