The measure of aspect ratio in ciou loss

Zzh-tju / CIoU

Distance-IoU Loss: Faster and Better Learning for Bounding Box Regression (AAAI 2020)

GNU General Public License v3.0

319 stars 44 forks source link

Closed John-Yao closed 3 years ago

John-Yao commented 3 years ago

Hi,

I am curious about the aspect ratio term of ciou, why you use arctan and don't use w/h directly?

Have you try w/h + l2 loss directly? Could you provide some details about it?

Thanks!

Zzh-tju commented 3 years ago

w/h is unbounded, so arctan() makes it bounded in (0, π/2)

John-Yao commented 3 years ago

Thanks!

383156098 commented 1 year ago

w/h is unbounded, so arctan() makes it bounded in (0, π/2)

为什么试用arctan约束在0~π/2，用tanh或者其他函数约束到0~1可以吗？