CS229-bagnet (Final Project)

Final project for CS229

• run_resnet50.py is the python script used to train, validate, and test the ResNet-50 baseline model.

• eval_resnet50.py is the python script used to evaluate the ResNet-50 baseline model on the test set.

• run_bagnet33.py is the python script used to train, validate, and test the BagNet-33 baseline model.

• eval_bagnet33.py is the python script used to evaluate the BagNet-33 baseline model on the test set.

• bagnet33_experiments.py is the python script used to run and evaluate the patch blackout experiments.

• data_preprocessing.ipynb includes the python code and outputs for data investigation and splitting for the flowers dataset.

• bagnet33_confmat.ipynb includes the python code and outputs for BagNet-33 evaluation, with a main focus on its confusion matrix.

• Performance of each model is stored in the model_performance_results directory, including loss_acc_plots, terminal output, and model checkpoints.

• The flowers_original directory contains the original downloaded flowers dataset, downloaded from Kaggle at: https://www.kaggle.com/alxmamaev/flowers-recognition.

• The flowers_tvtsplit directory contains the flowers data split into 70% training, 20% validation, and 10% test data subsets obtained by running data_preprocessing.ipynb.

• The paperwork directory contains the proposal, poster, final report and relevant figures for our CS229 project.

References:

• BagNets: "APPROXIMATING CNNS WITH BAG-OF-LOCAL FEATURES MODELS WORKS SURPRISINGLY WELL ON IMAGENET" - https://arxiv.org/pdf/1904.00760.pdf
• Datasets: https://www.kaggle.com/alxmamaev/flowers-recognition