I have multiple camera models with different intrinsics, how can I adjust SuGaR's code?

[yifanlu0227]

Hi @Anttwo ,

I notice that SuGaR assumes all cameras have the same intrinsics. However, my dataset includes multiple camera models. Could you give me some key points to adjust the current code? For example, I need to modify the render_image_gaussian_rasterizer function.

Thanks!

[yifanlu0227]

I've made it.

[bo233]

Hi there, I meet the same problem as you met. It seems that there are too many codes need to be modified. Can you show me of the places where you made modifications? I am extremely grateful.

[yifanlu0227]

Hi @bo233. The modification is straight forward.

1. replace those code like sugar.image_height to sugar.nerfmode.training_cameras.height[camera_indices.item()]. This allows you to use individual camera intrinsics stored in nerfmodel. You probably need to modify those files: sugar_extractors/coarse_mesh.py, sugar_extractors/refined_mesh.py, sugar_scene/sugar_model.py, sugar_trainers/coarse_density.py, sugar_trainers/coarse_sdf.py, sugar_trainers/refine.py

2. initialize new mesh rasterizer for each camera. Do not use the sole rasterizer. You probably need to modify those files: sugar_extractors/coarse_mesh.py, sugar_scene/sugar_model.py

---

My modificiation:

:100644 100644 e19668e c738f97 M        sugar_extractors/coarse_mesh.py
:100644 100644 eb4a815 be75956 M        sugar_extractors/refined_mesh.py
:100644 100644 9ae5947 bb479f6 M        sugar_scene/cameras.py
:100644 100644 9cee6ba d293411 M        sugar_scene/sugar_model.py
:100644 100644 a8a4a1f d95d4ad M        sugar_trainers/coarse_density.py
:100644 100644 cf0ef41 00bea09 M        sugar_trainers/coarse_sdf.py
:100644 100644 f5e17bb 7431548 M        sugar_trainers/refine.py