Details about the match_cost and loss weight?

[exiawsh]

Hello, thanks for your great work! I am trying to reproduce your paper. But I got bad results on azimuth and size. But under the supervision of Cartesian coordinate system, these two results are normal. I set both match_costs and loss weights of size and azimuth 1.0 follow detr3d and petr. In polar coordinates， I set the polar ray match_costs and loss weights 20.0 follow your paper, and I got normal results on map and mATE, but the results of size and azimuth is bad. So what about the details setting about the two param when using Hungrain match?

[outsidercsy]

Thanks for the attentin to our work. Some settings are incorrect. Object's position is parameterized by radial distance $r$ and azimuth $alpha$ $(sin(alpha), cos(alpha))$. Only the weight of $alpha$ $(sin(alpha), cos(alpha))$ is set to 20.0. The weights of radial distance, size and orientation are set to 1.0. The weights of polar velocity components (v_rad, v_tan) are set to 0.2, the same as DETR3D.