Can TripoSR work on x-ray datasets?

[pILLOW-1]

Hi, great work!
I have some interesting observations. My dataset contains pseudo-colored x-ray security images which have different prohibited items. I wonder if TripoSR can reconstruct the prohibited item given an x-ray image containing it. So I tried, and the results are as follows:

The first example is bullet, and the second one is a knife. It seems like TripoSR cannot easily reconstruct an x-ray item, and you can see obvious structure distortions. My question is: How can we improve TripoSR to make it work on x-ray datasets?

[koiker]

You probably need to train a model with thousands of X-ray images labeled. (Already identified what is each object and his 3D representation).
When you have that, the model will correctly reconstruct similar objects.