KinfuTracker camera pose to instant-ngp

[frjaa]

Hello everyone!
For context: I've acquired a series of depth and color images from the same object, using a RealSense camera, by moving the camera to another angle, close to each other, trying to keep the camera at the same height. The colmap2nerf.py script from instant-ngp creates a very good "transforms.json" that, when used in instant-ngp itself, is able to reconstruct a very nice 3D representation from those pictures.
Now I'm trying to use KinfuTracker to register the depth images and provide the camera poses, so I can build the "transforms.json" for instant-ngp.
But I still have had no success.
Does anyone have any hints? I understand that the Y and Z axes have opposite directions between Kinfu and instant-ngp, but there might be some more details I couldn't find. I've provided the camera intrinsic (fl-x, fl-y, cx, cy), an idea of grid length and grid center, and the depth images. Am I on the right track?
Thanks a lot!