DQEN: Dual Query Enhancement Network for DETR-Based HOI Detection

This paper presents DQEN, a method for improving HOI detection (identifying what interactions are happening between humans and objects in images) by enhancing the queries used in DETR-based models (a type of deep learning framework that treats object detection as a sequence prediction problem). The key innovation is using better initial information for object and interaction queries instead of random starting values, which helps the model recognize humans, objects, and their interactions more accurately.