Running on LLFF

[Srinivas-512]

Hi! I'm trying to run InstantSplat on LLFF data to compare with other models. I had a couple of queries regarding this:

• I modified dataset_readers.py file to align with the llffhold method of choosing train and test images (readColmapSceneInfo2). I also edited the helper file used by init_geo.py to choose the same images. However, I am running into an issue of no test poses. Also save_pose() in train.py isnt working as all the colmap camera ids are the same. Could you please advise me on what I need to do to run on LLFF?
• The number of initialization points from the MAST3R model is very high, much more than what COLMAP generates. Is this normal?

I'm using n_views = 3 , llffhold = 8

Thanks in advance!