Automated training of a convolutional neural network on the example of the problem of aircraft classification by image using tensorflow The repository is a set of files with classes for automatic training and configuration files for them (config folder). Basic entities:
- base - folder with abstract classes for datagenerator, model and trainer
- dataloaders - folder with data generators. Regulates the process of loading data, their preprocessing and determines in what form they will be submitted to the model
- models - folder with models classes (CNO) and setup for them (transfer models, optimizers, regularizers)
- trainers - folder with trainers for models. The trainer parameters regulate the training process of the model (fine-tuning stages, selection of callbacks, loss function, etc.)
- callbacks - folder with users callbacks
- plot - folder with user function for vizualization
- utils - folder with modules to help the pipeline
- config - folder with all settings for all part of pipeline
File main.py - entry point to the pipeline. Combines all parts of the pipeline into a single whole