This repository contains an implementation of a CNN model using a similar MobileNet architecture
Since I haven't messed around with the hyperparameters enough, the best model currently starts overfitting at around 15 epochs which isn't too bad far a smaller model but could definetly be improved with some better data preprocessing.
pip install -r requirements.txtSince this is a notebook, I advise you install a jupyter lab server on your own (https://jupyterlab.readthedocs.io/en/stable/getting_started/installation.html)
The training and testing is done in a single function train_and_validation_loop with the given number of epochs, device to run on and log writer for Tensorboard display.
The best models (with lowering evaluation loss) are saved in the saved_models dir and Tensorboard logs are stored in the runs dir.
Currently the model achieves around ~76% accuracy without too much overfitting.
What is nice with this architecture is obviously the reduced number of parameters, which is ideal for smaller models needing to run on mobile systems (about only a third of the number of trainable parameters than with the original bottleneck architecture using the same input data !)
Model paramters:
Training Loss:
Train VS Validation loss:
Since I haven't tested this notebook much, some variables like local paths may need to be changed to accomodate your needs.