How to get the correct bbox.txt intrinsic.txt ?

[dedoogong]

I reviewed related issues here but there is still no correct way found.
Can you please provide your code to get the values?
I'm struggling to reproduce your result with the provided images(especially real, big size objects using colmap. not small, synthetic ones from blender).
I've always failed to get the reasonable rendering results, just seen so ambiguous images.

for example, in Wineholder case,
colmap shows reasonable result so that I can trust it.

(below is Fountain case,

But the bbox.txt using "https://github.com/yxie20/lego_nsvf/tree/master/poses" is :
-1.7405309358237018 0.8937903721375522 0.8837398050743352 2.0868807956960875 3.9037096850320445 3.3750371294069716 0.3239107072481569

and your provided bbox.txt is

-0.5884649 -0.44142154 -0.12279411 0.78653513 0.43357844 0.8772059 0.125.

it's totally different!

camera intrinsics are little similar

888.0 0.0 400.0 0.0
0.0 888.0 400.0 0.0
0.0 0.0 1.0 0.0
0.0 0.0 0.0 1.0

VS

875.000000 400.000000 400.000000 0.
0. 0. 0.
0.
1.
800 800
~
Thank you!

[MultiPath]

The bounding box is pretty empirical to decide.

The bbox obtained by "https://github.com/yxie20/lego_nsvf/tree/master/poses' are not compatible with our code setup

[MultiPath]

The easiest way to decide is based on cameras.

[dedoogong]

I see, but can you please make a simple example(how you decide the bbox/voxel values on what criteria) for the case of Barn or fountain? I'm really struggling hard to make use of your work~! your work is almost only way to produce the nice voxels(it can be converted to mesh nicely, that's what I want)

[dedoogong]

This is my voxel (I defined it) and pcd(got from colmap). Im stilll getting not that good enough result

and the rendered voxel seems still not aligned to the object pose yet(below voxels correnpond to above rendered objects)

[dedoogong]

for better comparison, I captured again

After loading the pcd model in open3d or other tools, I found the object is not loacted in the origin but quite biased (ex, T': (150, 180, 20)). Also the object was rotated along all axis(eg., R' : x-axis=45, y-axis=10, z-axis=90) Do I need to move pcd and rotate the pcd to fit into the "straight-up" pose by R', T'(and every camera poses too by R',T')?