Argoverse to KITTI conversion : Bounding Boxes offsetted in Ground Truth lidar BEV

[NaveenPrasaad]

We are trying to benchmark the PIXOR detection results in paper "Train in Germany, Test in The USA: Making 3D Object Detectors Generalize" where we tried converting from Argoverse dataset to KITTI dataset.
We used argo2kitti.py from https://github.com/cxy1997/3D_adapt_auto_driving/blob/master/convert/argo2kitti.py#L105

Our code uses only one camera 'ring_front_center' . However when I check the conversion result, we observed a shift in the bounding boxes in BEV . I made sure that I didn't change any math in the bounding boxes coordinates conversion .
With the converted Argoverse dataset in KITTI format, calculated mAP is only 0.40, which is significantly lower than expected.

We would appreciate any assistance in resolving this issue. Thank you in advance for your help.