Description

How to Train D-Fine Object Detection on a Custom Dataset notebook added

D-FINE is a powerful real-time object detector that redefines the bounding box regression task in DETRs as Fine-grained Distribution Refinement (FDR) and introduces Global Optimal Localization Self-Distillation (GO-LSD), achieving outstanding performance without introducing additional inference and training costs.

D-FINE is available in 5 different sizes, ranging from 4M to 62M parameters, and capable of achieving from 42.8 to 55.8 mAP on the COCO dataset. It is also available in Object365+COCO trained 4 different sizes, ranging from 10M to 62M parameters, and capable of achieving from 50.7 to 59.3 mAP on the Object365 finetuned models