GQA-SGL

New framework enhances compositional VQA with disentangled concepts

Researchers have introduced a new framework called Disentanglement-based Equivariant Learning (DEAL) to improve compositional visual question answering (VQA). This approach uses causality-inspired interventions to disen…