Weak-to-Strong Compositional Learning from Generative Models for Language-based Object Detection