Zhixin Zhang

Thanks for your interest. It's hard to say how well the model will work in the zero-shot scene, and we didn't test that. Maybe it's better to train on your...

Are you meeting some problems when deploying the pretrained weights? There will be many factors to affect the performance. For example, 1) how is the image data **normalized and preprocessed**?...

Thanks for your interest. Our experiments were held on 2D BEV coordinates and there might be no height information. If you want to integrate it into your 3d model, changes...

Yes, ego. The final prediction is in the ego coordinate system.

Maybe there is something wrong with your vis. code. This vis. tool from [another repo](https://github.com/hustvl/MapTR/blob/maptrv2/tools/maptrv2/av2_vis_pred.py) might be helpful (z is default set to 0). 

Is this on the nuScenes dataset? I only tried their code on Argoverse 2 and it worked well. Maybe there is something different between these two datasets.

Seemingly you just use the "lidar2img" transformation. However, there might be some augmentation (scaling, typically) that will introduce additional transformation. This should be processed like this: ``` for lidar2img, img_aug...