How to use 4×4 RT matrices from RealEstate10K in ViewCrafter inference?

[chenjiahes]

Thank you for your great work on ViewCrafter!

We are currently trying to evaluate ViewCrafter on the RealEstate10K dataset as a baseline in our experiments. Specifically, we aim to generate novel views starting from a single image, following a trajectory of target camera poses provided in the dataset.

However, we encountered an issue:
RealEstate10K provides 4×4 camera extrinsic matrices (RT), while the infer.py script in ViewCrafter seems to take rotation angles (Euler or similar) as input.

Could you kindly clarify:

How can we convert the RT matrices to the input format expected by infer.py?

Or is there an existing interface/script in the codebase that directly supports RT input?

We would greatly appreciate your guidance, as we would like to include ViewCrafter in our comparison fairly and correctly.

[chenjiahes]

Sorry for the bother — this is quite important for our ongoing evaluation work.

[nnanhuang]

Hi, I am also want to use RT matrices as input, have you figure it out?

[mo230761]

Hello, I am also want to learn, have you figure it out?