Multiview feature fusion in nuscenes: camera pose

[trayOWO]

Hi, I'm trying to do multi-view fusion 2D fusion on occ3d-nuscene dataset, generating point query(with coordinate -1~1) as lidar point reference.

When preprocessed 2d data, I found that there's no data file "pose" for camera pose parameter(4 x 4), so I used extrinsic: translation(3) and rotation(4) to do multiview fusion, but it resulted that no any point can be projected on image(dim: 800 x 450).

May I ask how to get the pose data(4 x 4) for nuscenes image?

Note: I found it's different between occ3d-nuscene's frame token and nuscene_2d you provide 's frame token, so I couldn't use the camera parameter from nuscene_2d directly.

Thanks.

[stickygao]

i am doing the same thing， do you solve it?