Giovani in un'ora - Ciclo di seminari - Quarta parte

Day - Time: 20 November 2024, h.11:00

Place: Area della Ricerca CNR di Pisa - Room: C-29

Speakers

• Luca Ciampi
• Chiara Mannari
• Giuliano Cornacchia

Referent

Fabio Carrara

Abstract

Abstract: "Object counting estimates the number of objects in images or video frames. Studies reveal that the human brain employs two distinct methods for counting objects owing to the subitizing ability: the efficient Parallel Individuation System (PIS) for fewer than five objects and the less accurate Approximate Number System (ANS) for larger quantities. Thus, automated counting is undoubtedly helpful (at least) in dense scenes, making it a key area in computer vision with wide applications.

Recently, research has shifted towards class-agnostic counting (CAC), which counts objects of arbitrary categories not seen during training, unlike traditional class-specific methods that require separate models for each object type (e.g., people or vehicles). CAC draws inspiration from humans' instinct to discern what merits counting when confronted with unfamiliar objects, allowing users to specify object categories at inference time without retraining. With advances in vision-language models, prompt-based CAC now allows natural language input for specifying object classes.

However, we identify significant limitations in current benchmarks for evaluating this task, which hinder accurate assessment and the development of more effective solutions. Specifically, we argue that the current evaluation protocols do not measure the ability of the model to understand which object has to be counted. This is due to two main factors: (i) the shortcomings of CAC datasets, which primarily consist of images containing objects from a single class, and (ii) the limitations of current counting performance evaluators, which are based on traditional class-specific counting and focus solely on counting errors. To fill this gap, we introduce the Prompt-Aware Counting (PrACo) benchmark, which comprises two targeted tests, each accompanied by appropriate evaluation metrics. We evaluate state-of-the-art methods and demonstrate that, although some achieve impressive results on standard class-specific counting metrics, they exhibit a significant deficiency in understanding the input prompt, indicating the need for more careful training procedures or revised designs."

Chiara Mannari - "End-User Modelling as an Approach to Link AI-Driven Hints with Computational Thinking Skills"

Abstract: "Computational thinking (CT) is a way of formulating problems and solutions originating from computer science. This cognitive process has broad applications, enabling individuals to manage complexity in various domains. Recent studies formulated CT in terms of a set of skills - abstraction, decomposition, algorithmic design, evaluation and generalisation - and introduced tools to assess and train these skills. Our approach aims to combine these efforts with AI-driven hints, a technique to guide users through context-aware suggestions, personalised feedback, and adaptive support.

We explore the possibility of leveraging modelling as an approach that activates CT skills. In doing so, we evaluate a block-based modelling tool for end-users by setting up a user study under the framework of CT skills.

In this seminar, we present the design of our exploratory study."

Abstract: "Urban transportation systems often struggle to meet local mobility demands and manage traffic effectively. Alternative Routing (AR) algorithms show promise by dispersing traffic across multiple routes for common origin-destination (OD) pairs, reducing congestion. However, AR effectiveness is tied to road network characteristics. We introduce DiverCity, a metric for assessing OD pair diversifiability based on alternative route geography. Analyzing 23 global cities, we found DiverCity increases with distance from city centers, longer OD distances, and wider OD angles. High DiverCity correlates with reduced traffic concentration, suggesting it as a potential proxy for congestion. To explore network structure impacts, we designed a toy model with a grid and ring road, showing that speed limits affect DiverCity: as ring-road speed rises, route diversity falls. Fine-tuning speed limits can enhance diversity without new infrastructure. Real-world tests in cities with ring roads (e.g., Rome, Berlin) confirm these patterns, highlighting the value of strategic speed adjustments. This work provides insights into urban network design, supporting city planners in enhancing route diversity and managing traffic."