Unsupervised Domain Adaptation through rotation as regression for RGB-D Object Recognition

Final project of the course Machine Learning and Deep Learning at Politecnico di Torino developed by me and my colleague Marco Gullotto, MarcoChain. The aim of the project is to replicate some experiments of Loghmani et al. paper (https://arxiv.org/pdf/2004.10016.pdf) and to implement two new variations of the algorithm. The repository includes the following files:

• presentation to have a quick overview of the project;
• Demp.py contains the CNN's architecture;
• Syn_vision.py contains the classes used for creating loaders;
• Accuracy.py contains the functions used for testing the models;
• Functions.py contains a miscellaneus of classes and functions: class to compute cross entropy and function to visualize the images;
• main.ipynb contains all the implementations of our experiments.
• project_mldl.pdf: final essay.