Moderating Illicit Online Image Promotion for Unsafe User-Generated Content Games Using Large Vision

Link: https://arxiv.org/abs/2403.18957

Conference: Usenix Security 2024

Keywords: Vision-Language Models, Online Safety\

Threat Model

Sexually explicit and violent user-generated content games (UGCGs)

Method

Vision-Language Models (VLMs) to detect UGCGs

alt text

alt text