c2w incorrect for test images in instant-ngp-data data parser?

[hoanhle]

Describe the bug When running

ns-train nerfacto --output-dir /home/leh19/workspace/experiment/outputs/base_line --vis viewer+tensorboard  --experiment-name painting_1 --pipeline.model.background_color "random" instant-ngp-data --data /home/leh19/test_run_1/JPG/painting_1/processed 

I got the shape mismatching issue.

   assert c2w.shape == num_rays_shape + (3, 4), f"Shape mismatch: {c2w.shape} != {num_rays_shape + (3, 4)}"
AssertionError: Shape mismatch: torch.Size([4096, 4, 3]) != torch.Size([4096, 3, 4])

And this issue only happens when running with instant-ngp-data data-parser. Using nerfstudio-data data-parser was okay.

Data to reproduce: https://drive.google.com/drive/folders/1LlWcbTCr2EO9g-VhNowhW1IRiMHbLzgZ?usp=drive_link

[nepfaff]

This seems weird as it worked for my data in instantNGP format. Are you sure that your data is formatted correctly (I didn't look it your data myself)? An easy test would be to see whether instantNGP takes your data. If it does, then this is indeed a bug.

Extrinsic matrices are always a pain as there are too many different standards. I would need to double-check which ones are used by Nerfstudio and instantNGP respectively.