How to create target files for xBD Dataset for BDA Task? Questions About MambaBDA

[applerengg]

Hello,

I am very interested in this work and I'm currently trying to experiment with the MambaBDA model.
The model requires images/ and targets/ directories for the data. However, targets for tier3 data are not included in the official xview2 downloads. Also tier3 is the largest part so I think it is important to include in the training.

My questions are:

1. How did you create the targets for tier3? is there any script to create the targets in the xBD dataset or in changemamba repo (I couldn't find any)?
2. If not, what is the correct method to generate target files? (I know there are label.json files but not sure how to correctly utilize them.)

Additional questions;
3) the xview2 provides the dataset in forms or .tif and .png files. As far as I understand, these datasets are not any different (exact same amount of images). I assume the paper used png variation. Is there any importance on which one to use, does the model work with .tif files?
4) Are you going to share MambaBDA xBD trained checkpoints?

Looking forward to your answers, best wishes.

[applerengg]

I created a nice script that creates the target masks and also checks the error rates, comparing with ground truth. So questions 1 & 2 are resolved. I am planning to share the code in my related public repo if anyone's interested.
For question 3, I decided it's not important.
Question 4 still stands.