euiyeon-kim / dlow-pytorch Goto Github PK

I want to know the license of this project.
Can I refer to this code to modify the dlow model？

Thanks for the nice reproduction code, I have some questions about training details:

• Number of the source training set. Did you use the entire(24,966 images) GTA-5 data set for training.
• Number of epochs. Did you use the default num_epochs = 1000?

I try to train it with RTX-3090, and set batch_size = 1. 1000 iterations take about 35 minutes. If using the entire GTA-5 and setting num_epochs = 1000, it seems unlikely because it would take a lot of time. Can you share with me the training details？

• Implement Discriminator
• Implement Generator
• Implement Optimizer

1. Have to experiment whether to use Identity loss or not
2. Have to test CycleGAN is working well

• Model saving and loading
• Logging + Visdom

• DataSet
• DataLoader
• Data spliter