Scientific seminars given by experts working in industrial sectors on topics such as medical and biomedical imaging, barcode reading, autonomous driving, cultural heritage, ...
Two-dimensional barcodes (such as QR and Datamatrix) are widely used in warehouse logistics and high-speed production pipelines to automate product tracking. However, to handle various size packages, it happens frequently that small and high-resolution barcodes are challenging to decode. Conventional solutions address such challenges by utilizing expensive hardware (e.g. CPU, FPGA, ASIC) or powerful lighting sources, increasing the costs of the system. This study introduces a multi-step, scalable, and adaptive super-resolution (SR) method that focuses primarily on the areas where barcodes are present and minimizes the computational burden on the uniform regions of the background. Our approach achieves superior image quality by dynamically determining the required refinement steps for each region of the image analyzed. Experiments demonstrate that our method outperforms the state-of-the-art SR models on barcode images, reaching higher PSNR and decoding rates while reducing the latency.
Vision Transformers (ViT) are very powerful deep architectures capable of recognizing and using very subtle statistical patterns, often invisible to humans. Nice but... to achieve these performances they require having large databases, well maintained and representative of the domain of interest, perhaps even i.i.d.. This usually only happens in paper-academic contexts or technological superpowers. But fortunately (for you) in this seminar we will show how it is possible to profitably use ViTs in real low-budget industrial contexts where the raw material (i.e. data) is few, dirty and often ugly.
This seminar will cover some concepts and recent advances in the emerging field of self-supervised learning methods for solving imaging inverse problems with deep neural networks. Self-supervised learning is a fundamental tool deploying deep learning solutions in scientific and medical imaging applications where obtaining a large dataset of ground-truth images is very expensive or impossible. The seminar will present different self-supervised methods, discuss their theoretical underpinnings and present practical self-supervised imaging applications. Finally, I will discuss my experience developing and collaborating on open-source software for science (https://deepinv.github.io/), and some of the lessons learned along the way.
Computer vision has evolved from a niche into, I believe, the most extensively investigated field in all of computer science. The relatively recent and ever-increasing adoption of deep neural network has fostered most of the key advances in computer vision research, while enabling a breadth of novel and diverse application, many of which were simply unthinkable just a few years ago. Within this complex and rapidly evolving landscape, I will present on overview of computer vision applications, from classical ones in the space of manufacturing to image and video generators alongside Multi-modal LLMs than can reason about images. The second part of the talk will provide a deeper view on research, with a focus on some projects currently going on at CVLab which address topics related to manufacturing, healthcare and autonomous driving. Finally, I will offer a different research perspective pertaining neural networks seen as data that can be processed by other neural networks to perform classical computer vision task and also interact with LLMs. This is the very recent research line of meta-networks and neural fields, which CVLab has contributed to pioneer in the last three years.
The Indy Autonomous Challenge and the Abu Dhabi Autonomous Racing League represent two of the world’s most groundbreaking competitions in autonomous racing. Nearly ten teams from across the globe compete for multimillion-dollar prizes, showcasing autonomous vehicles capable of racing at ever-increasing speeds. The events feature both head-to-head and multi-vehicle scenarios—with three or more cars simultaneously on the track—pushing the boundaries of artificial intelligence and engineering.
Unimore Racing, representing the University of Modena and Reggio Emilia (UNIMORE), has consistently ranked among the top three teams in each competition. Notably, they secured second place in a head-to-head race in Indianapolis, reaching speeds exceeding 290 kph, and claimed victory in Las Vegas during the first four-team autonomous race in history. This ongoing success highlights Unimore Racing’s expertise and the continued evolution of autonomous driving technologies.