I start this blog to share research publications that is interesting to follow. The selection is based on my previous work and my previous solutions on this topics. It is always good to see how the new technologies can be applied to resolve them in a better way.

Finally a good UV atlas solution!

When we work on room creation project, we will need to create texture for light map backing. We also need to create Gen 3D model with UV atlas. In both these cases, we want a clear texture with small number of parts, clear, semantic meaningful UV layout. Traditional UV unwrapping does not respact the part semantic, which leads to UV space fixing really difficult.

PartUV could be particularly valuable if you:

• deal with AI-generated meshes (which tend to be messy, noisy, or non-optimal),
• need a robust UV unwrapping solution that avoids many tiny charts,
• want semantic / part-based UV organization (per-part atlases, per-part packing),
• care about automation (parallelization, handling degenerate meshes),
• need a UV unwrapping method that more gracefully handles real-world / noisy geometry rather than pristine artist-made meshes.

PartUV: Part-Based UV Unwrapping of 3D Meshes

Triangle is the future!

The idea to use triangle as representation of gaussian splat and rebuild the foundation from training to rendering is a great idea. The utilization of the primitive of hardware accelectrated graphics component as the representation of the splats gives a large boost of the rendering speed. The author has two research work this year, and the second one even construct the triangle patch so well that it can directly used in rendering engine without any plugins. This is the paper I do want to implement and try myself. I think it has much better application comparing to the original representation of gaussian splats.

Triangle Splatting+: Differentiable Rendering with Opaque Triangles