I am currently running inference with pretrained models on a dataset from my university. For the different cameras and datasets, there are varying cam2img/cam2ego matrices, which makes sense. However, I've noticed that the cam2img matrices in the NuScenes dataset are simple 3x3 matrices, whereas those in the KITTI dataset are 4x4 matrices. In the case of KITTI, the first 3x3 matrix represents the intrinsic camera parameters, if I'm not mistaken. Unfortunately, I can't find any documentation on the composition of these matrices, and having that information would be very helpful for me to adapt them to my own dataset.
Suggest a potential alternative/fix
Short overview over the composition of the different cam2img matrices and how they are generated.
Branch
dev-1.x branch https://mmdetection3d.readthedocs.io/en/dev-1.x/
📚 The doc issue
I am currently running inference with pretrained models on a dataset from my university. For the different cameras and datasets, there are varying cam2img/cam2ego matrices, which makes sense. However, I've noticed that the cam2img matrices in the NuScenes dataset are simple 3x3 matrices, whereas those in the KITTI dataset are 4x4 matrices. In the case of KITTI, the first 3x3 matrix represents the intrinsic camera parameters, if I'm not mistaken. Unfortunately, I can't find any documentation on the composition of these matrices, and having that information would be very helpful for me to adapt them to my own dataset.
Suggest a potential alternative/fix
Short overview over the composition of the different cam2img matrices and how they are generated.