some issues about code of detect

[yangyang567]

HELLO, I'm really interested in your brilliant work but i still got a few questions, the code below:

dx1 = (y[..., 0] * 2) 2 s_gain dy1 = (y[..., 1] 2) 2 s_gain dx2 = (y[..., 2] 2) 2 s_gain dy2 = (y[..., 3] 2) 2 s_gain y[..., 0] = (x_coord.view(bs, self.na, ny, nx)+1 - (dx1)) self.stride[i] y[..., 1] = (y_coord.view(bs, self.na, ny, nx)+1 - (dy1)) self.stride[i] y[..., 2] = (x_coord.view(bs, self.na, ny, nx) + (dx2)) self.stride[i] y[..., 3] = (y_coord.view(bs, self.na, ny, nx) + (dy2)) * self.stride[i] xyxy = y[..., :4].view(-1,4)

In the code,the pred y is the format of (xyxy) which means the top-left postion and the bottom-right postion.But in the paper it should be the format of (LTRB) as below. So do I misunderstand it or can you tell you about the code of the prediction (LTRB)? Thank you.

[yangyang567]

ok,i know it