DeepLearningCV is a project exploring different ways to use deep learning for image classification.
Exploring different approaches for solving a computer vision with deep learning, and solving issues that occur when implementing the resulting model on resource constrained edge devices.
My intent for the notebook is to see what kind of accuracy can be achieved by a small model, then see what improvement can be found using a pretrained transfer learning approach, and finally training the small model with a different approach called knowledge distillation using the transfer learning model to train the small model along with the dataset.
The project structure is missing a few things from the .gitignore file. To reproduce locally you will need to add the Kaggle dataset to the project so that it looks like this.
. deep-learning-cv
. - data
. . - train
. . - valid
. - notebooks
Once you have the data folder in the project things should work fine. I used the models folder as a storage location for models that I wanted to keep a copy of, as the ones generated by the notebook are put in the notebooks folder for now.