DETR-based algorithms
From Wikipedia, the free encyclopedia
DETR-based (DEtection TRansformer) algorithms are a family of object detection algorithms that involve transformers to identify and locate objects in images. Their fundamental concepts were introduced in the seminal 2020 paper by Carion et al..[1][2]
DETR-based algorithms treat object detection as set prediction, where predictions correspond to a subset of learned queries (candidate objects). A fundamental training component is a loss that involves bipartite matching of predicted and ground-truth objects.
