Writing a SIGGRAPH paper from scratch
Web page · https://alexandergoslin.me/writing-a-siggraph-paper-from-scratch
First seen by Oriel on .
AI summary
The author recounts independently developing Terrain Diffusion and InfiniteDiffusion with one RTX 3090 Ti, turning the project into a paper accepted at SIGGRAPH 2026.
- The project aimed to create a learned procedural terrain generator that integrates easily with game engines while supporting infinite, deterministic, nonperiodic, and randomly accessible worlds.
- The author built a machine-learning foundation through several years of textbooks, courses, and papers, recommending surveys and recent work before recursively following references.
- Infinite generation requires independently generated regions to agree at their boundaries, unlike ordinary image patches whose padding and differing context can produce mismatched edges.
- The proposed convolutional approach removes padding and supplies extra input context, making overlapping output pixels receive identical inputs and therefore align by construction.
- Padding-free diffusion models remain challenging because they exclude architectures using nonlocal operations such as attention and incur substantial computational costs in downsampling UNets.
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Points
Score reported by the source.
- Lobsters: 8, latest observation ; +7 over observed period since
Comments
Comment count reported by the source.
- Lobsters: 0, latest observation ; 0 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Lobsters | 793ecg | https://alexandergoslin.me/writing-a-siggraph-paper-from-scratch/ Read discussion |