Expect dw to bring more gain in yolov6！

[meituan-gengyifei]

hi, we are interested in your work, and you are welcome to add dw's work to our yolov6 for even greater gains
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]

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

[justin14567]

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