Modifying camera pose for converting to image dataset

[alafumee]

Hi, I am trying to obtain a dataset with images from another camera pov by replaying the provided playdata/demo.hdf5 dataset through dataset_states_to_obs.py. According to information from the robosuite documentation, I should be able to do this by modifying the agentview camera params in _setup_camera() in LIBERO/libero/libero/envs/problems/libero_kitchen_tabletop_manipulation.py. It doesn't seem to work as the camera view stays the same when I visualize the converted dataset with playback_robomimic_dataset.py. However, after I modified the said function, the rollout videos saved while training the low-level policy do follow the new camera configurations.

Does anyone have more insight into this or tried to do so themselves? Thanks a lot!