Camera Parameters for SEER Datasets

[jo1jun]

Hello, thank you for the great work and for releasing the SEER model!

I’m trying to explore different training setups using SEER, and I have a question regarding the datasets used (CALVIN, LIBERO, and DROID):

Is it possible to obtain the camera intrinsic and extrinsic parameters for both the base-view and gripper cameras for each image frame in these datasets?

If such information is not explicitly logged, would it be reasonable to approximate the gripper camera’s extrinsic parameters using the robot state (i.e., the position and orientation of the end-effector)?

Thank you very much!

[Nimolty]

Dear jo1jun,
The camera intrinsics and extrinsics in CALVIN and LIBERO are unchanged, which can be found in the official github. Perhaps it is not easy to found since it is hidden in deep directories. The camera intrinsics and extrinsics in droid can be acquired in droid raw dataset.