You are reading the documentation for the in-development version of VSGAN.

Models#

Model files are PyTorch .pth files with trained network parameters stored within. They are the same as state dictionaries. Think of it as configuration of a Network based on the training it’s had.

VSGAN must support the model file’s specific architecture for the model file to work. A list of Supported Models are listed on the homepage.

Where can I get Model files?#

Know any more places to find trained models? Let me know!

Do I need to train a Model file?#

Yes and no. VSGAN is not an ESRGAN model trainer, it’s a model tester. You will need to provide your own trained model file for VSGAN to use. If you do not have a Model file, you can either find one trained by someone else, or train your own.

How can I train my own Model file?#

You may wish to train your own model file for various reasons. E.g., the model files you find don’t work for your scenario, or you feel like you could get a better result.

There’s a few things you must realise about training a model.

  1. It takes a lot of patience, and constant running of your GPU hardware at high-usage.

  2. It takes a lot of effort to create the dataset that will be used to train the model how you like.

  3. A lot of storage may be used for the dataset, and using an SSD (NVMe if possible) is recommended.

  4. Writing a lot to an SSD (or NVMe) can be dangerous to it’s lifespan if you constantly write a lot of data.

  5. Among all of this, there’s quite a lot more to learn than what you would learn to use VSGAN. There will be a lot of words like Losses, LR, HR, GT, Validation, Act, Norm, Layer, Arch, and so on.

If you still feel like you want to give it a try, go for it! Check out traiNNer it’s a PyTorch training and testing toolbox for various Model Architectures. For instructions on using traiNNer, check their readme. VSGAN will not provide help for training.