The repository contains the code for Project 1 of the Deep Learning (EE-559) course at EPFL during Spring term 2021.
This project is accomplished by:
- Riccardo Cadei: @riccardocadei
- Riccardo Fradiani: @riccardofradiani
- Niccolò Polvani: @nickpolvani
In this study we compare different Deep Neural Networks to predict the inequality among 2 gray-scale images representing handwritten digits from MNIST database. Advantages of weight-sharing and auxiliary loss are also discussed. Training the models on a training set of 1 000 couples of images we got a test error rate equal to 2.96%.
For a detailed description of our solution, read report.pdf
The project has been developed and test with python3.8.3.
Required libraries: Pytorch, matplotlib.