Machine Learning
CNN for Templates Matching
Use of Mask R-CNN for Templates matching thanks to data augmentation and transfer learning. The system can detect an object in complex situations given only a single image.
Research, implementation and analysis of the most suitable machine learning models for 3D object classification. Development of a proof of concept of an automated workflow that recognises features in traditional CAD models and enhances them with essential metadata.
Winner of PREMIOSWISSENGINEERINGTICINO USI Master Informatics 2021 (best Master thesis)
Use of Mask R-CNN for Templates matching thanks to data augmentation and transfer learning. The system can detect an object in complex situations given only a single image.
ROS implementation that use a CNN regressor to predict the angular velocity of the Mighty Thymio to avoid obstacles and pitfalls.
Naive Implementation of Procedural Voronoi Foams for Additive Manufacturing which generates foams with different elasticities.
Implementation of a Neural Machine Translation using an encoder-decoder model with attention mechanism for phrase-based translation.
People counter and noise detection with Arduino nano 33 BLE thanks to a CNN for objects detection and classification.
Game with random generated maze, special textures and two types of game views, normal and isometric. The maze is floated with water cycles and the player need to collect as many coins as possible.
Research of which extent off-the-shelf Internet of Things module can be programmed to function as an HSM and development of a proof of concept with it.
Design and development of a little scenario with animated objects, moving camera and day/night light simulation. The scene have reflections, shadows and textures.
Using a long short-term memory network to predict the next textual character given a sequence of characters to generate chunks of text.
Pre-processing and analysis a data-set of demographic characteristics of people to build a model to forecast the income.
This is an application to manage bookmarks and share them with other users. The backend is Laravel and the frontend is React. The authentication is performed with Laravel passport and API.
Development of an efficient classifier with selective search and Bag of Visual Words to determine the presence or absence of logos within a given set of logos.