search

strongwolf / DW

A Dual Weighting Label Assignment Scheme for Object Detection
Apache License 2.0
135 stars 17 forks source link

Expect dw to bring more gain in yolov6！ #16

Open meituan-gengyifei opened 2 years ago

meituan-gengyifei commented 2 years ago

hi, we are interested in your work, and you are welcome to add dw's work to our yolov6 for even greater gainshttps://github.com/meituan/YOLOv6 We actually tried it on yolov6n using the dw you open sourced on fcos. But the effect is not ideal. When the box-refine branch is used, it will drop by 1.1map, and when the box-refine branch is not used, it will drop by 1.6map. This may be due to your special design for the fcos network, or the yolov6n network is too lightweight. So I expect you to introduce more targeted dw on yolov6 to improve the effect.

strongwolf commented 2 years ago

Thanks for your interests in our work. We will try to apply dw on yolov6 in the coming days.

justin14567 commented 1 year ago

Hi, What do you use to draw the Sample Weighting diagram?