TUM/ICL Poses

[sakatoable]

❓ Questions and Help

Hi,

I am wondering how the TUM poses are loaded from the dataloader.

Suppose I have specified to get a sequence length of 3 images, i was expecting to get 2 poses for 3 images such that:
pose_1 = image_1 -> image_2
pose_2 = image_2 -> image_3

But, I noticed that the data loader is returning 3 poses, so what does the third pose correspond with? Am I correct to assume that pose_1 and pose_2 belong to the above mentioned sequence?

[Nik-V9]

Hello, Thank you for reaching out with your query. The TUM data loader returns poses, which are relatively inverse transformed where the first pose would be identity (I) and the other poses in the sequence would be relative to the identity pose.

For example, in your case, the following would be the case:
pose_1 = Identity
pose_2 = relative transform of image_2 with respect to image_1
so on ...

That's why you will get 3 poses for 3 images in your sequence.

[krrish94]

Closing due to inactivity