Jurgen Gurakuqi's repositories
deep-learning-based-dog-breed-classifier
The goal of the project is to improve a kaggle project about Dog Breed Classification, achieving an higher test accuracy. The original project achieved 79% of accuracy on the test set, while this one goes up to 87%. Also further improvements were made to the data processing pipeline in terms of modularity and performance.
graph-kernels-and-manifold-svm
This project aims to compare the performance obtained using a linear Support Vector Machine model whose data was first processed through a Shortest Path kernel with the same SVM, this time with data also processed by two alternative Manifold Learning techniques: Isomap and Spectral Embedding.
parallel-rasterizer
The goal of this project is to implement a parallel 3D software rendering pipeline with programmable fragment shader.
search-based-sudoku-solver
This project aims to show how to solve a given Sudoku in two different ways, through Backtracking (in a flavour of Forward Checking) and Relaxation Labeling.
ai-powered-web-library
The goal here is to develop a simple mock web app book library for demonstrating a possible use of AI for aiding users
ar-cube-projection
The goal of this project is to capture the pose of an object given its coordinates, to be able to project a virtual cube over each frame of the video.
foreground-background-segmentation
The goal of this project is to perform the segmentation of some specific videos to separate background and foreground, extracting the silhouette of the object in the video, frame by frame, producing a video with the processed frames.
marker-detection-and-tracking
The goal of this project is to capture the position of the concave corner of each numbered marker per each frame of the video, and store all the information in an output csv file.
mitsuba-snapshot-tool
The goal of this project is to develop a powerful and user-friendly tool that allows users to produce a dataset of synthetic images for the purpose of testing Shape from Polarization methods, and even further shape reconstruction techniques.
pattern-estimation-and-trend-analysis
The goal of these examples is to analyse the given datasets to determine whether some models can be established for purposes of prediction, to assess how stepwise prediction behaves with respect to a personally chosen model and determine an unknown trend in the cereal dataset.
ranker-comparator
The goal of this project it to provide a tool to build new ranker easily and to compare them with existing ones in terms of results overlapping.
sfp-methods-comparison-extended
The goal of this project is to extend the project from Smith et al. by fixing some critical errors and bottlenecks, and introducing multithreading for the processing of multiple SfP scenes.
slt-based-spam-filters
This project aims to develop three spam filters using different Machine Learning techniques. The techniques to be used are Support Vector Machines with Linear, Polynomial and Radial Basis Function kernels (together with their angular versions), Naive Bayes and K-Nearest Neighbours.