Training

The training submodule gather the different functions required to encode (i.e. train a model on) an image.

• warmup.py is about comparing a list of different random initialization and selecting the best one after a few hundred training iterations

• train.py is the actual training loop used to perform a number of optimization iterations.

• loss.py measures the performance of the model to enable learning.

• test.py measures more quantities than loss does, e.g. the neural networks rate

• quantizemodel.py is called at the end of the training to quantize the neural networks rate

• preset.py contains training recipe with hyper-parameters controlling the training process.

• loss
• train
• test
• warmup
• quantizemodel
• preset