| - 2019-03-14 1.9.0 beta: added deep features from VGG19 and a discriminator for GAN training. |
| Moved all non strictly architecture building operations outside of the model files. |
| The models are combined when needed in the Trainer class. |
| In order to allow for GAN training `fit_generator` function had to be replaced |
| with the more granular `train_on_batch`. Now the project relies on |
| custom data handlers and loggers instead of the custom Keras generator. |
|
|
| - 1.9.2 beta: Included tutorial notebooks, modified init files for less verbose import statements. |
| Comprehensive docstrings refactor. |
|
|
| - 2019-03-25 1.9.3 beta: now ISR models are child of the ImageModel class, whose purpose is to collect |
| common functions across models. At the moment it only contains a predict function that takes numpy arrays |
| as input and returns numpy arrays in a proper image format. |
|
|
| - 2019-03-26 2.0.0 beta: brought a number of training hyperparameters to the surface of trainer and the CLI assistant. |
| Removed most hard-coded variables and improved assistant. Weights saving now allows metric selection, with suggested metrics list in assistant. |
|
|
| - 2019-04-01 2.0.1 beta: Cleaner training session tracking: now weights are saved together with a yaml configuration file that contains hyperparameters and other relevant training details. |
| Partial migration to pathlib from os. |
|
|
| - 2019-04-02 2.0.2 beta: Predictor uses the model parameters and the configuration file rather than inferring details from the name of the weights. Minor changes to input/output folder structure. Partial migration to pathlib. |
| Added configuration settings to pre-trained weights. |
|
|
| - 2019-04-03 2.0.3 beta: Parametric metrics and losses in trainer. Custom weights initialization range with RandomUniform initializer. |
| Added PSNR evaluation on Y channel (for literature comparison). Automatic session config generation from trainer input parameters. |
| Uniform naming for feature extractor and name change for generator network. |
|
|
| - 2019-04-16 2.0.5 beta: Added flatness check scheduler. Refactored some trainer variables into dictionaries for a more compact configuration summary. |
|
|
| -2019-05-30 2.1 beta: Added large image inference. Minor fixes and added new non-artifact-removing GANS model. |
|
|
