hustvl / MapTR

[ICLR'23 Spotlight] MapTR: Structured Modeling and Learning for Online Vectorized HD Map Construction

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Which coordinates does maptr choose?

ailias opened this issue · comments

In the code here , it announced using LIDAR coordinates, but LIDAR coordinates x is front, see bellow figure, so if we specific (bev_h, bev_w) = (200, 100), the point_cloud_range should be xbound =[-30, 30, 0.15], ybound=[-15, 15, 0.15], zbound=[-10, 10, 20].
image

Actually, the code is xbound =[-15, 15, 0.15], ybound=[-30, 30, 0.15], zbound=[-10, 10, 20], and then exchange LSSTransform's bev feature h and w to match the (bev_h, bev_w) size.

So I am confused about the coordinates. Can you help me understand it? Thank you.

In the nuScenes dataset, the x-axis is not forward.