This project was made in collaboration with Christina Grand and Laura Fur in our course Deep Learning for Visual Recognition. Our project can be read here. In this project, we explored different ways to quantify uncertainty for deep learning models such as MC dropout and Bayesian Neural Networks.
Our main .ipynb file where the code for the project is done is at Project_DL.ipynb.
We also have another .ipynb where we for example did some experiments on our CNN model is at Project_Step_by_step_and_gradCAM.ipynb.
To run the notebook download the dataset German traffic sign benchmark dataset from Kaggle. Upload the zip file to your Google Drive. I chose the path German-traffic-signs/archive.zip, so if you change the location change the corresponding line in Python. The images we used for out-of-dist images can be found here.