The code in this file develops a flower image classifier across 102 different types of flowers. Current implementation includes the following:
- Option to train the model on GPU using Google Colab
- Tracking of metrics for train and validation sets
- Saving of checkpoints and best model
- Transfer learning based on pre-trained models to speed up training time
- Adjustment of learning rates over time
- Graphical display of output
Planned optimizations:
- Identification of optimal learning rate
- Loading last checkpoint to re-start training
So far the best model has achieved a 95% accuracy on the validation set in less than 10 epochs of training.
- In training on Google colab it was noticed that with poor internet connection or during peak periods model training performed poorly. If accuracy rates seem low re-try with better connection or during lower rates of usage.