Image-Text-Emotion Dataset Curator/Annotator
Image
Emotion RecognitionThis project involved curating and labeling a large-scale image-text-emotion dataset for a vision-language generation framework. I manually labeled and augmented 25,000 image-text pairs across seven emotion classes to enhance model controllability. Data quality was ensured using LLM-based augmentation and RoBERTa-based automatic filtering to optimize the labeled dataset for downstream evaluation and training. • Labeled images according to specific emotional classes for multimodal model adaptation. • Combined manual data annotation with AI-based enhancements for comprehensive dataset creation. • Implemented quality control using advanced NLP methods and filters. • The labeled dataset was integral to training and evaluating controllable vision-language AI systems.