search

xinshuoweng / AB3DMOT

(IROS 2020, ECCVW 2020) Official Python Implementation for "3D Multi-Object Tracking: A Baseline and New Evaluation Metrics"
http://www.xinshuoweng.com/
Other
1.67k stars 404 forks source link

Question about the KITTI detections. #110

Closed chenxyyy closed 2 years ago

chenxyyy commented 2 years ago

Hi! I have try your project on KITTI-tracking datasets. But I found that the detection results offered in your project is a little strange. The loction of z can reach 50m. This is almost impossible in reality

https://github.com/xinshuoweng/AB3DMOT/blob/933e4af2ef4c04a7c970f9abf8ae0dc2739ab77e/data/KITTI/detection/pointrcnn_Car_val/0001.txt#L5

I think the definition of each column is as follows. If I understand correctly, how should the numerical value z explain?

  1. frame
  2. type
  3. box_2d_0
  4. box_2d_1
  5. box_2d_2
  6. box_2d_3
  7. (I don't know)
  8. h
  9. w
  10. l
  11. x
  12. y
  13. z
  14. ry
  15. orientation
vkorotkine commented 1 year ago

The reference frame is such that z points forward, y down, x right. So z=50m is very plausible.

PleasantPhoenix commented 9 months ago

参考系是这样的：z 指向前方，y 向下，x 向右。所以 z=50m 是非常合理的。

您好，能简单说明一下您的这个文件中每个数字的含义吗？