Non Convergence on a scene

[Fortunanto]

Hi, I've been trying to train NR-Nerf on my synthetic scene, that looks like so:

I'm using the multi-view setup, and I've set the poses to be the poses calculated by colmap, and for some reason, it's failing to converge, even after tweaking the hyper-parameters very heavily

The results themselves are essentially a black screen, so the net seems to converge on a black image for some reason.

any idea what might be the cause for the issue?

Thanks!

edit: the background of the scene is transparent.