PointGT: Simultaneous Geometry and Texture Editing for Point-Based Representations

PointGT introduces a point-based 3D representation built for simultaneous geometry and appearance editing. That matters because many modern reconstruction and view-synthesis pipelines can produce convincing assets, but those assets are still awkward to modify once captured or generated.

The core idea is to combine a point-based structure that handles deformations well with a learned UV mapping approach for high-resolution texture edits. In practice, that means the same asset can be reshaped and retextured without the usual split between geometry tools and appearance tools. The authors say the method supports fine-grained edits while keeping rendering quality high.

This is especially relevant for workflows built around 3D Gaussian Splatting and other neural representations, where texture-editing approaches have often been incompatible with geometry changes. If the technique holds up outside controlled demos, it could make neural asset iteration more usable for production-style pipelines, especially for teams that want fast capture plus later art direction.

The paper was submitted on 3 September 2026 and sits at the intersection of computer vision and graphics. For game developers, the practical question is whether point-based neural assets can become editable enough to fit real content pipelines, rather than staying as attractive but brittle research outputs.