Unable to generate model_tsfm and model_tsfm_semantic files for a synthetic dataset.

[s58vshar]

Hi,

I am trying to run the method on a dataset that I create from blender. I have attached a rendered image for reference. I suspect it has to do with lack of iris in the rendered image. Is there any way to convert the bust object I have already to model_tsfm or any other suggestions how I could make it work?

Thanks in advance

[0mil]

@s58vshar Hi,

What exactly does it mean to run the method on data created in Blender? As far as I understand, if you want to run this program using blender model image instead of real world videos, I believe that it might be impossible. What I mean is that the essential mechanism of this program called neural radiance fields fundamentally depends on real world light information.

Best regards,

[KeyuWu-CS]

@s58vshar If you use blender, I think you can get correct camera pose with multi-view images. In such case, you don't need to run COLMAP and directly use the correct camera parameters run Instant-NGP. For model_tsfm, you can slightly modify the code and not use iris to optimization. Another possible method is to directly use your head model(include scalp) and set an identity matrix to save it in model_tsdf.

[s58vshar]

Hi @KeyuWu-CS ,

Yes I have the camera parameters without colmap. My issue was with bust_fitting. I have been trying to use my own head model and to extract the scalp but I havent succeeded yet. Thanks for your response nonetheless 👍