chriskeraly

As far as I understand, this seems impossible since if stride =1 I directly get sparse submanifold behavior? Is there a work around?

Any chance you could add a license file?

There is an inconsistency in this repository of how yaw is calculated. As a consequence the anchors in motionformer are usually oriented wrong. See screenshot bellow, which has the main...

There is a missing transpose [here](https://github.com/OpenDriveLab/UniAD/blob/main/projects/mmdet3d_plugin/datasets/nuscenes_e2e_dataset.py#L506) ```python l2g_r_mat = l2e_r_mat.T @ e2g_r_mat.T ``` should be ```python l2g_r_mat = (l2e_r_mat.T @ e2g_r_mat.T).T ``` This has consequences when updating the reference points...