General PrinciplesΒΆ
Generally there are two things you can do with SimpleSR. You can either train models on a dataset, or you can evaluate already trained models on your test data.
For training you will need to:
Obtain a set of training images (links to some popular Super-Resolution data sets can be found here.
Initialize a
DataPipelinewith your training dataInitialize an
SRModel- TheSRModelneeds aGenerator- this the component that learns to upscale images - Optionally you can also provide aDiscriminatorto theSRModelto train in adversarial/GAN modefinally you need to start the training process with the initiaized
DataPipelineandSRModel
For evaluation you will need:
A
DataPipelineinitialized with your test images (initialized withevaluate_only=True)The path to a saved model
See the examples folder for some training and evaluation examples and read the docs for training/evaluating models for further info.