Open-Vocabulary Scene Graph Generation as Proxy Task for Child Sexual Abuse Material Detection

• Grant number: 24/21679-6
• Support Opportunities: Scholarships abroad - Research Internship - Scientific Initiation
• Start date: November 20, 2025
• End date: February 19, 2026
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Sandra Eliza Fontes de Avila
• Grantee: Artur Alves Cavalcante de Barros
• Supervisor: Jefersson A dos Santos
• Host Institution: Instituto de Computação (IC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
• Institution abroad: University of Sheffield, England
• Associated to the scholarship: 24/09372-2 - Scene Graph Generation as Proxy Task for Child Sexual Abuse Material Detection, BP.IC

Abstract

Child sexual abuse materials (CSAM) have reached alarming proportions in the digital age. According to the National Center for Missing and Exploited Children's CyberTipline, over 36 million reports of suspected CSAM were received in 2023, making it a record year. The availability and sharing of such harmful content online not only exacerbate the trauma inflicted upon the victims but also significantly burden law-enforcement agents who have to inspect thousands of files, leading to emotional strain manually. In light of that, there is a need for reliable automated tools that can handle this kind of material securely and efficiently. We aim to design, develop, and deploy a solution based on machine learning to detect CSAM, supporting forensic analysis automatically. This undergraduate research project, which is related to the research internship abroad through the ``Bolsa Estágio de Pesquisa no Exterior'' (BEPE) program by FAPESP, investigates how scene recognition and representation can be explored for CSAM detection while respecting the limitations of working with CSAM (i.e., low-data regimes and restricted testing). We aim to explore how to employ Machine Learning techniques for Scene Graph Generation (SGG) to create helpful scene representations for the CSAM detection task. More specifically, by incorporating the flourishing Open-vocabulary SGG task, we seek to enhance the model's ability to handle diverse and unseen scenarios, paving the way for more robust and scalable CSAM detection solutions.