I have a question about the camera-controlled version. How do you process the camera trajectories of the dataset to solve the scale-invariant problem? I am planning to finetune your camera-controlled model, so I need to process my dataset to align with yours.
