Too Long; Didn't Read

Our method uses a probing-based approach to gather critical image details essential for detecting hateful content in memes. We design VQA questions for this purpose, generate Pro-Cap as image captions, and integrate them with BERT-based and PromptHate models for comparison in hateful meme detection effectiveness.