Workshop Introduction

 Title
Theoretical Results on Viewing Graphs

Abstract
Structure from Motion (SfM) is a fundamental task in Computer Vision that aims at recovering both cameras and the 3D scene starting from multiple images. The problem can be conveniently represented as a “viewing graph”: each node corresponds to a camera/image and an edge is present between two nodes if the fundamental (or essential) matrix is available. While several research efforts on SfM have focused on devising accurate and efficient algorithms, much less attention has been devoted to investigate theoretical aspects. In particular, a relevant question is establishing whether a viewing graph is “solvable”, i.e, it uniquely determines a configuration of cameras. This talk will give an overview on existing formulations and different notions of viewing graph solvability.

 Title
Vocabulary-free Image Classification

Abstract
Image classification is one of the most traditional computer vision problems, aiming to assign a semantic label to images. Standard approaches for this task require a labeled training set and a list of target class names available beforehand. This talk will show how recent vision language models can be re-purposed to sidestep both needs. Specifically, we will talk about how contrastive vision-language models coupled with a large textual dataset allow us to perform this task without training and without needing a list of classes beforehand. We will conclude the talk by linking this result with the emerging vision-by-language paradigm, generalizing these large models beyond the scope they were designed for.

 Title
Innovation in Computer Vision: from Feature-Based to Deep Learning. VRAI Research Group’s journey between Past, Present and Future.

 Title
Indoor AGV navigation and localization for logistics using both traditional computer vision and modern approaches

Abstract
In the last years, starting from the rise of Industry 4.0 and thanks to the recent advances in the widespread of intelligent systems that enable real-time decision-making processes, new paradigms and concepts regarding to the autonomous robot navigation and localization have been established, especially in the context of logistics and transport. With reference to the manufacturing industry, a new generation of autonomous robots (AGV – Automated Guided Vehicle) can be effectively used to move payloads (either heavy or not) within a warehouse and to enable autonomous and intelligent navigation via smart interaction with unstructured environments, people and/or other things. In this talk different AGV modules designed and developed using either computer vision and image processing techniques or deep learning models is discussed. This case study is an example of integration of both traditional approaches as well as AI-based solutions. The aim of the talk is to open the discussion about the opportunity of using one approach instead of the other, based on the specific domain requirements.

Round Table and Discussion