Methods and references

1Reconstruction from images

Image to 3D runs TRELLIS.2 on our own GPU in Vienna. The image is cut out with U²-Net (rembg), encoded with DINOv3 and decoded into a sparse voxel structure, then our post-processing (below) turns it into one watertight solid in millimetres. Text to 3D first draws the object with SDXL.

1. J. Xiang, X. Chen, S. Xu, R. Wang, Z. Lv, Y. Deng, H. Zhu, Y. Dong, H. Zhao, N. J. Yuan, J. Yang. Native and Compact Structured Latents for 3D Generation (TRELLIS.2). arXiv preprint 2512.14692, 2025. preprintmicrosoft/TRELLIS.2 · MIT · Image to 3D
2. J. Xiang, Z. Lv, S. Xu, Y. Deng, R. Wang, B. Zhang, D. Chen, X. Tong, J. Yang. Structured 3D Latents for Scalable and Versatile 3D Generation. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), highlight, 2025.microsoft/TRELLIS · MIT · Image to 3D
3. O. Siméoni, H. V. Vo, M. Seitzer, F. Baldassarre, M. Oquab, et al. DINOv3. arXiv preprint 2508.10104, 2025. preprintfacebookresearch/dinov3 · DINOv3 License (commercial use with conditions) · Image to 3D
4. X. Qin, Z. Zhang, C. Huang, M. Dehghan, O. R. Zaiane, M. Jagersand. U²-Net: Going deeper with nested U-structure for salient object detection. Pattern Recognition 106, 107404, 2020.xuebinqin/U-2-Net · Apache 2.0 (rembg: MIT) · Image to 3D
5. D. Podell, Z. English, K. Lacey, A. Blattmann, T. Dockhorn, J. Müller, J. Penna, R. Rombach. SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis. International Conference on Learning Representations (ICLR), 2024.Stability-AI/generative-models · CreativeML OpenRAIL++-M · Text to 3D

2Depth, tactile reliefs and images

The photo relief tool runs Depth Anything V2 Small in your browser through transformers.js: the image of a painting or an object never leaves your computer. Tactile reliefs for blind and partially sighted visitors follow the layered approach of Reichinger et al. Images can be enlarged with Real-ESRGAN before a lithophane or a relief.

6. L. Yang, B. Kang, Z. Huang, Z. Zhao, X. Xu, J. Feng, H. Zhao. Depth Anything V2. Advances in Neural Information Processing Systems (NeurIPS), 2024.DepthAnything/Depth-Anything-V2 · Apache 2.0 (Small model) · Photo relief
7. A. Reichinger, M. Maierhofer, W. Purgathofer. High-quality tactile paintings. ACM Journal on Computing and Cultural Heritage 4(2), 5, 2011. · Photo relief
8. X. Wang, L. Xie, C. Dong, Y. Shan. Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data. IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), 2021.xinntao/Real-ESRGAN · BSD-3-Clause · Lithophane

3Watertight solids, repair and simplification

Generated and scanned surfaces are often open or self-intersecting. We rebuild a solid volumetrically, deciding inside and outside by ray parity voted over nine directions as in Nooruddin and Turk, extract the surface with marching cubes and simplify it with quadric error metrics, preserving topology.

9. F. S. Nooruddin, G. Turk. Simplification and repair of polygonal models using volumetric techniques. IEEE Transactions on Visualization and Computer Graphics 9(2), 191–205, 2003. · GPU repair
10. W. E. Lorensen, H. E. Cline. Marching cubes: A high resolution 3D surface construction algorithm. ACM SIGGRAPH Computer Graphics 21(4), 163–169, 1987.scikit-image/scikit-image · BSD-3-Clause · GPU repair
11. M. Garland, P. S. Heckbert. Surface simplification using quadric error metrics. Proceedings of SIGGRAPH ’97, 209–216, 1997.cnr-isti-vclab/PyMeshLab · GPL-3.0 (server only) · GPU repair

4Fabrication and cultural heritage

Splitting a large model into printable parts with connectors is the problem studied by Chopper. SplitSTL offers a grid partition and a natural-joint partition that cuts each part alone at its narrowest neck, with pins or magnets and engraved part numbers. The heritage literature below frames why replicas and tactile models matter in museums. The photogrammetry tool (AliceVision) is in development.

12. L. Luo, I. Baran, S. Rusinkiewicz, W. Matusik. Chopper: Partitioning models into 3D-printable parts. ACM Transactions on Graphics 31(6), 129 (SIGGRAPH Asia), 2012. · Splitter
13. R. Scopigno, P. Cignoni, N. Pietroni, M. Callieri, M. Dellepiane. Digital Fabrication Techniques for Cultural Heritage: A Survey. Computer Graphics Forum 36(1), 6–21, 2017. · Splitter
14. M. Neumüller, A. Reichinger, F. Rist, C. Kern. 3D Printing for Cultural Heritage: Preservation, Accessibility, Research and Education. 3D Research Challenges in Cultural Heritage, LNCS 8355, 119–134, Springer, 2014. · Splitter
15. C. Griwodz, S. Gasparini, L. Calvet, P. Gurdjos, F. Castan, B. Maujean, G. De Lillo, Y. Lanthony. AliceVision Meshroom: An open-source 3D reconstruction pipeline. Proceedings of the 12th ACM Multimedia Systems Conference (MMSys), 241–247, 2021.alicevision/Meshroom · MPL 2.0

5Open-source software

Libraries that run in SplitSTL without a paper of their own.

6Open data

The open collection indexes the 3D scans published by the Statens Museum for Kunst (SMK), Copenhagen, through its open API, under the Public Domain Mark 1.0. Titles were translated by us; dimensions and licences come from the museum’s records.

7Licences of the models we run

Every model on the paid path allows commercial use. We keep a written audit of each licence and refuse non-commercial components (NVIDIA nvdiffrast, MeshLib, Depth Anything V2 Base/Large) at run time. Details are on the About page.

How to cite SplitSTL

SplitSTL: an open 3D lab for cultural heritage. M. Buessler, Vienna, 2026. https://splitstl.com