Multimodal AI Safety Dataset (Personal Project)
Curated and annotated image-text pairs for toxicity detection in memes using an annotation workflow and defined safety objectives. Exported labeled data into standard formats for downstream dataset usage and analysis. Applied annotation consistency practices to support AI safety dataset quality. • Curated 500 meme image-text pairs for toxicity detection • Labeled content using Label Studio • Exported annotations to COCO + CSV formats • Prepared dataset for multimodal safety evaluation