Dynamic MDETR: A Dynamic Multimodal Transformer Decoder for Visual Grounding.

Multimodal transformer exhibits high capacity and flexibility to align image and text for visual grounding. However, the existing encoder-only grounding framework (e.g.