object detection

LocateAnything speeds up vision-language grounding

LocateAnything uses parallel box decoding to make visual grounding faster while improving high-IoU localization quality.