Data preperation for Real World Scene

[Priyam1204]

I want to use Pixie for 3D reconstruction on my own images. I have images in nerf studio format, with transform.json converted from COLMAP, but when I looked at the real world scene directory, the format is very different. Can there be a guide to prepare data properly for running inference on Pixie?

[vlongle]

Can you share with me the data that you're trying to run our method on? I can take look at look later!

[Priyam1204]

I am using this dataset . So basically, I recorded a 32-second 360 video and broke it down to 250 frames and then used it for reconstruction.
Here are my main doubts:

1. Are there any methods to calculate object bounds and voxel size for scene_bounds.json?
2. Similarly, is there some method to prepare custom_object_config.json, or is it just hit-and-trial based?
3. I also tried reproducing the bonsai and vasedeck examples, and I found that the particles after voxelization were very high in my case (90K) compared to vasedeck (6k). I used the same bounds and voxel size as the Vasedeck example. Also, the Segmentation Accuracy was very low in my case compared to Vasedeck. Is this related to how the bounds and voxel size are set?
4. When I used the custom_vasedeck_config, I saw that the MPM solver particles were more than the Voxelization/material field confident particles. But in my case, this was the opposite, and the MPM solver particles were about 50% less. What is the cause of this?

Any insights or recommendations would be greatly appreciated. Thanks in advance for your help!