search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.58k stars 7.95k forks source link

CDIoU-CDIoUloss #7537

Open hujunchao opened 3 years ago

hujunchao commented 3 years ago

Can anyone use this mothed in YOLOV4? author says "CDIoU and CDIoU loss have different excellent performances in several models such as Faster R-CNN, YOLOv4, RetinaNet and ATSS. There is a maximum AP improvement of 1.9% and an average AP of 0.8% improvement on MS COCO dataset, compared to traditional evaluation-feedback modules. "

hujunchao commented 3 years ago

https://arxiv.org/pdf/2103.11696.pdf https://github.com/Alan-D-Chen/CDIoU-CDIoUloss/tree/5b911bc2cfe39c671a4e514d06ea73ad1f53292e

blake-espeland commented 3 years ago

You can use nms_kind=diounms in your config file for yolo layers. I don't believe diou loss is an option, but loss=ciou is a valid parameter in yolo layers as well.

In my experience, using those two in my cfg file increased my mAP (I'm not sure how much exactly).

akashAD98 commented 3 years ago

@blake-espeland have you tried nms_kind=greedynms , why diounms better than grredynms? can you please explain me.