Referring Expression Comprehension: Grounding natural language in visual scenes using Words as Classifiers approach

A modular multimodal framework that generates object masks from text prompts. It uses a lightweight cross-modal decoder to fuse features from interchangeable vision and language backbones.