Gianluca Scarpellini Email & Phone Number

New York, Ny, Us

Working on generative models for biotech applications.

My contributions involve• Multi-GPU utilities and unit tests• Instance segmentation for kaggle competitions• Instance segmentation mAP metric• Documentation for object-detection augmentations and metrics

Genoa, Liguria, Italy

Genoa, Liguria, Italy

• Contributed to a novel Diffusion Models formulation based on graphs that achieves state-of-the-art results on ordering text and outperforms long-lasting optimization methods for solving puzzles (over 98% accuracy on puzzles of any size) https://github.com/IIT-PAVIS/Positional_Diffusion/• Designed a pipeline that leverages curiosity-driven exploration to improve a robot object-detector by 6% in challenging photorealistic scenarios without any human interventions;• Devised the first pipeline for solving 3D human pose estimation with a single event camera, outperforming stereo-based approaches https://github.com/IIT-PAVIS/lifting_events_to_3d_hpe• Collaborated on proposing a novel method for person re-identification with event-cameras;• Mentored 3 master students working on Diffusion Models.

Genova, Liguria, Italia

• Developed an open-source framework for event-based vision https://github.com/IIT-PAVIS/event_library• Trained models for dynamic behavior classification from event-camera stream with precision/recall >95% (PyTorch)

Londra, Inghilterra, Regno Unito

• Conducted extensive research and developed a novel algorithm to optimize offline policy evaluation on real robots with visual foundation models, improving efficiency by up to 10x;• Adopted best practices for scaling jobs in a complex distributed environment;• Paper accepted at ICLR2024

Milano, Italia

• Directed a project on aerial semantic segmentation through deep learning, achieving IOU > 80% through hyperparameters optimization (Keras, Hyperopt);• Improved models for car model and maker classification (200 classes) with accuracy > 80% (Keras);• Engineered a plate detection & OCR pipeline on Raspberry PIs for real applications;

Bergamo

• Server developing (ASP.NET, Node.JS)• Database maintenance (SQLServer, MongoDB)• +2,000 users

Bachelor thesis working on YOLO for cars, pedestrians, and cyclists detection from RGBD data

Bergamo

M.S. In Computer Science, Computer Vision, 110L/110, Gpa 29.45/30

• Computer vision e Pattern Recognition (Point features, Stereo vision, Bayesian Filters) • Deep Learning (CNN, RNN) • Data & Text mining.

B.S. In Computer Science, 110L/110, Gpa 29.5/30

• Computer Vision and signal processing • Pattern Recognition • Machine Learning

Nanodegree, Deep Reinforcement Learning

I never stop learning and improving my skills with state-of-the-art tools. I'm taking the Deep Reinforcement Learning program to improve.

Sensor Fusion Nanodegree Program

I'm focused on robotics applications of computer vision and perception algorithms as part of my career as a computer vision engineer..