FlowCV Logo
resume profile picture
Steve ZARETTI R&D Engineer
[email protected]
+32 496 71 84 73
Rue de la descente, 31 - 6061 Charleroi
10 Nov 1992
Belgium
Single
A, B
steve.zaretti.be
GitHub
kossolax
Profile

Graduated with a Master's degree in Computer Science, I currently hold the position of a R&D Engineer at ACIC, while also serving as a professor in social promotion, teaching in the evening classes. My passion for programming was ignited during my young age, and I enjoy refining and sharing my technical skills every day.

Professional Experience
R&D Engineer, ACIC

As a software developer specializing in Artificial Intelligence at ACIC by Chapsvision, my main job has been to drive our tech forward, pushing the boundaries of what's possible in AI. I mainly focus on object detection, using Convolutional Neural Networks architectures to achieve the best results. With ACIC, I've been working on various collaborative Research & Development initiatives, including but not limited to:

Jan 2017 – present
  • RUDIS (2023 – ongoing): Our goal with RUDIS is to create energy-efficient models that reduce energy consumption. To achieve this, we've been exploring the best compromises between different frameworks, models, and devices. We started by optimizing the entire processing pipeline, from stream acquisition and image decoding to GPU throughput. Our efforts have paid off, as we've managed to reduce the overall cost of inference by a factor of three. We've also successfully trained a student model that matches the detection capabilities of its more complex teacher counterpart in a specific scenario, using distillation loss. The student model version has 6x fewer parameters, we are now trying to quantize it to reduce even more!
  • Steve ZARETTI
    1 / 5
  • ReconnAIssance (2022 – onging): In this project, we've been working on developing a system to detect lost objects in train stations. To achieve this, we've refined our understanding of crowd movement patterns by applying advanced calibration techniques and in-depth temporal analysis. We've also developed a method to identify ownership of items using a stable wedding algorithm, and to group people together using a dbscan algorithm. Additionally, we've created a system to find lost objects by combining bbox overlap information with spatio-temporal data. Our goal is to make this system efficient enough to be embedded on edge devices, and we reached TRL-8 on this technology.
  • IRIS (2019 – 2022): We developed a deep learning approach on thermal images to detect and track small objects (just 20 pixels high). To do this, we created a bi-spectral model and a custom training loss function. A major part of the IRIS project was adding a sense of time to our software, based on the results of our neural networks. This significantly improved our MOT tracker, making our AI-powered PeopleCounting solution work at TRL-9, which is better than the traditional version TRL-8 that were using CV approach. The bi-spectral model and the new tracker also helped our PanoramaDetection system.
  • MobiLearn (2017 – 2020): We explored how deep learning can improve smart city initiatives. We also developed a system to detect road lanes and curves using traditional computer vision methods. As part of MobiLearn, ACIC built its own dataset from public CCTV footage, focusing on road traffic. The project resulted in ACIC's first production-ready model and tracker, based on the YOLOv4-tiny architecture, and introduced our initial exploration of tracking multiple objects. This laid the foundation for the ACIC Traffic solution, although some work remains to achieve a high-quality, real-world implementation as we achieved TRL-6.
  • SmartCount (2016 – 2018): We modelized CCTV field of view and benchmarked Machine-Learning vs Deep-Learning approach to crowd numbering problem. During this project, we found that the ResNet model, which was developed during the project, performed better than traditional pattern matching in real-time scenarios. By the end of the project, we opted for YOLOv3 with automatized hyperparameters based on the camera's view and crowd density. The findings and results of this project led to the development of a new ACIC project called CrowdNumbering and considered as TRL-9.
  • Teacher, ISIPS & HEPH Condorcet

    In addition to my roles in the tech industry, I have found great reward in sharing my knowledge and experience as a Higher Education Instructor. I provide instruction exclusively during evening hours, accommodating students who are committed during conventional hours and providing an opportunity for continuous learning. This approach facilitates learning for individuals who are occupied during regular hours and are looking to further their education in the IT sector.

    Sep 2018 – present
    Steve ZARETTI
    2 / 5
  • One of my key offerings is a comprehensive course in Computer Network, mirroring the curriculum offered by the globally recognized CISCO CCNA certification. This course provides both a theoretical understanding and practical skills in the realm of networking.
  • In addition, I conduct a course on Linux System Administration, guiding students to become proficient in managing Linux-based systems. This course digs into essential aspects of Linux administration.
  • Occasionally, I also give some teach sessions on Algorithms and Object-Oriented Programming. These classes take into core programming concepts and techniques, assisting students in sharpening their understanding the logic behind software development.
  • Software developer internship, M.H. Consultants

    During my internship, I built a form generator for giving psychological tests. This tool was made using ASP.NET and JavaScript. Some form was dynamic to detect attention issues. I worked on all parts of this project, from discussing the initial needs with the client to carrying out the final tests.

    2015
    Indie game developer

    This was my first big project where I've created a role-playing modification in the game Counter-Strike in SourcePawn. The full project contains more than 100K lines of code, and more than 30K different players were actively playing this mod!

    2014 – 2017

    I also took on the role of the main system administrator, hosting online sessions of this mod. My core duties included devising and implementing measures to mitigate the repercussions of network attacks, wich helped the game kept running smoothly and without interruption.

    Organizations
    CSMF, Board Member, CE/CPPT President

    Beyond my academic and professional commitments, I have been privileged to serve as a member of the Board of Directors at a local educational institution, which is notably where I received my education. In this strategic role, I am directly involved in guiding substantial investments and shaping the pedagogical trajectory of the school. As part of the board, we continuously evaluate our academic offerings, invest in impactful projects, and ensure our decisions align with our commitment to providing quality education.

    Jun 2019 – present
    Steve ZARETTI
    3 / 5

    Further expanding on my leadership roles, I preside over both the Works Council (CE) and the Committee for Prevention and Protection at Work (CPPT). As the President of these committees, my mission is to maintain a robust and constructive social dialogue between the employee's delegation and the management committee. Balancing the interests of these groups, I facilitate an open dialogue to ensure a harmonious work environment. By listening, mediating, and promoting consensus, I strive to keep our institution a productive, safe, and satisfying place to work and learn.

    Skills
    CAINSI, C99|C++C++03, C++11, boost|Deep LearningYOLOv2-v7, custom loss, custom data augmentation|Computer VisionOpenCV|LinuxDebian, CentOS, Docker|Network & SecurityCCNA, SE-Linux, iptables|PythonPython 3.7|JavaScriptTypeScript, Angular
    Languages
    French

    Native

    English

    B2

    Publications
    Education
    Sep 2015 – Jun 2018
    Steve ZARETTI
    4 / 5
    Interests
    Board games
    Video Games
    New space
    Steve ZARETTI
    5 / 5