some questions about RDIoU calculation. F(Θ(o), Θ(t), k, k)

[lejk8104]

Hello, thanks for your great work, RDIoU

I have some questions about RDIoU calculation.

RDIoU parameter setting

mentioned paper, Θ(o) = sin(Θ(gt))cos(Θ(a)), Θ(t) = sin(Θ(a))cos(Θ(gt)), k=1(default)

RDIoU Θ opereation

F(Θ(o), Θ(t), k, k) = min(Θ(o) +1/2, Θ(t) +1/2) - max(Θ(o) -1/2 + Θ(t) -1/2) => if ( Θ(o) < Θ(t) ) => Θ(o) +1/2 - Θ(t) - 1/2 => Θ(o) - Θ(t) = sin(Θ(g))cos(Θ(a)) - sin(Θ(a))cos(Θ(g)) = sin( Θ(g) - Θ(a) ) => else ( Θ(o) > Θ(t) ) => Θ(t) +1/2 - Θ(o) - 1/2 => Θ(t)- Θ(o) = sin(Θ(a))cos(Θ(g) - cos(Θ(g)sin(Θ(a)) = sin( Θ(a) - Θ(g) )

thus, F(Θ(o), Θ(t), k, k) = sin( Θ(a) - Θ(g) ) (if Θ(o) > Θ(t) ) , sin( Θ(g) - Θ(a) ) ( if Θ(o) < Θ(t) ) 1) it is RIoU work same operation (Θ), right? 2) I think, the operation of RIoU minimized difference that degree(=yaw) of rotated gt boxes and rotated predicted boxes. right?

[hlsheng1]

Hi, sorry i cannot understand your question 1, but your understanding on question 2 is correct.