Poses are camera-to-world?

[FreemanG]

According to your data description:

Poses are camera-to-world, not world-to-camera transformations.

Then why does normalize_cam_dict.py save W2C in the cam_dict? Shouldn't it be C2W (cam to world)?

[Kai-46]

You are right! You could get C2W easily by taking the inverse of W2C, and vice versa.

[FreemanG]

So I have to change this line

return np.linalg.inv(C2W)

to

return C2W

Is this right? The experiments seem better so far.

[Kai-46]

Yes, or you could do such matrix inverses while converting the json camera file to text files.