Alapfogalmak
The author argues that leveraging self-supervised learning is crucial for scene recognition in child sexual abuse imagery to combat online offenses effectively.
Kivonat
The content discusses the importance of automated tools in combating child sexual abuse imagery, highlighting the challenges faced by investigation centers. It introduces a novel approach of Indoor Scene Recognition and evaluates different models on publicly available datasets and CSAI directly. The study emphasizes the significance of self-supervised learning in producing powerful representations for downstream tasks, showcasing improved performance compared to fully supervised versions.
Statisztikák
Over 10 million child sexual abuse reports are submitted to the US National Center for Missing & Exploited Children every year.
Self-supervised deep learning models pre-trained on scene-centric data can reach 71.6% balanced accuracy on indoor scene classification task.
The study includes 615 random samples from CSAI dataset, with 313 images categorized as CSAI and 302 as Suspected CSAI.
Idézetek
"Crime in the 21st century is split into a virtual and real world." - Pedro H. V. Valois et al.
"The scarcity and limitations of working with child sexual abuse images lead to self-supervised learning." - Pedro H. V. Valois et al.