Repository for the MATLAB code for the final project of the UniTS course in Computer Vision and Pattern Recognition, a.y. 2021-2022. The project consists in a convnet multiclass classifier design and test.
The main branch of this repo includes three MATLAB codes, one for each task. In the assignement there were also short optional tasks that are included in the relative files. The folder 'workspace' includes MATLAB variables that needs long computational times, and so are saved separately for faster loading. One can optionally delete and recompute them to fit necessities.
The dataset used is the 'Fifteen scene categories' introduced in (Lazebnik, 2006).
The project was carried on by Damiano Ravalico and Lorenzo Elia.
Update 02/02/2022: The file CVPR_Final_project_3 had an error in the SVM formulation with Gaussian kernel. The 'KernelScale' parameter for the 'fitcsvm' function was missing and thus the test accuracy with this model was about 7%. Corrected and now is about 86%.