GSoC 2024 – Contributor Application

[Davide Macario]

Dear OpenCV Team,

I am Davide Macario, an Italian Master's student in Computer Engineering and Telecommunications from Politecnico di Torino. I am currently attending my last year of university in Chicago, at UIC, where I am working on my Master's thesis, "A Model-Distributed Inference Approach for Large Language Models at the Edge."

I have taken many university courses on machine learning techniques, computer vision, and AI. I have also developed a personal interest in these topics, leading me to work on hobby projects in this field. I am also currently enrolled in the OpenCV University course "Computer Vision and Deep Learning Applications", where I'm learning advanced computer vision techniques that I did not have the opportunity to learn in university.

Some recent projects I am most proud of include:

• My master's thesis (https://github.com/davmacario/MDI-LLM), where I distributed a Large Language Model (GPT-2) on a network of Nvidia Jetson boards to perform inference on memory-constrained, low-power devices
• "FREISA" (https://github.com/B-AROL-O/FREISA), the winning project of the OpenCV AI Competition 2023, consisting of a four-legged robot equipped with an OAK-D Lite camera and a water sprinkler, able to run computer vision models to identify plants and decide whether to water them or not depending on the leaves' condition; my main contribution to this project was the computer vision module (vision pipeline based on YOLOv8 running on the OAK-D camera).

I want to express my interest in taking part in Google Summer of Code, and I would be interested to contribute to either of the following projects:

• Quantized models for OpenCV Model Zoo
• Multi-camera calibration toolbox

I believe I have a deep enough knowledge of Python and a good foundation working with the most common libraries and frameworks for data science and ML, such as Numpy, Pandas, PyTorch, and OpenCV, paired with hands-on experience in the form of multiple projects both related to the university and not.
Furthermore, in both the projects I mentioned previously, I had to deal with the memory limitations of resource-constrained devices, and recently, while working on my thesis, I learned about the most common weight quantization techniques for ML models (neural networks specifically).
Additionally, my academic background taught me to understand and design mathematical models applied to computer vision, such as pattern recognition, landmark detection, and depth estimation. I believe these are essential components when working on the multi-camera calibration toolbox.

I have attached my CV, and you can find me on GitHub at https://github.com/davmacario.

I look forward to hearing back from you!

Davide