Research map / Point clouds / Shape and surface completion
RoofDiffusion: Constructing Roofs from Severely Corrupted Point Data via Diffusion
Kyle Shih-Huang Lo, Jorg Peters, Eric Spellman
This paper tried a common change: More diffusion sampling steps
increase inference steps from 60 to 500 · PoznanRD s99 i80 · RMSE (m): 1.123 -> 0.893 Improved
What 457 accepted papers report for this change
- 277 improved
- 77 no clear effect
- 26 mixed
- 77 hurt
- Optimality of FSQ Tokens for Continuous Diffusion for Categorical Data with Application to Text-to-Speech ICML 2026 · increase denoising steps from 5 to 25 for FSQ-original · SEED-TTS test-en · WER: 2.39% -> 2.00%Improved
- DiscoForcing: A Unified Framework for Real-Time Audio-Driven Character Control with Diffusion Forcing ICML 2026 · increase denoising steps from 10 to 100 · AIST++ · latency (ms/frame): 26.26 -> 261.91Hurt
- Variable-Length Tokenization via Learnable Global Merging for Diffusion Transformers ICML 2026 · more diffusion sampling steps (25 to 100) for our method · ImageNet 256x256, 32 tokens · gFID: 2.89 -> 2.54Improved
See this change in detailMatched by the wording of each paper's own result, read by AI from the full text. Papers report gains more often than failures, so this shows what gets published, not the chance a change works.
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use RoofDiffusion (footprint-guided) vs best interpolation baseline on PoznanRD with high sparsity and incompleteness | height map completion · PoznanRD s99 i80 | MAE (m): 0.827 -> 0.603 | Improved |
| use RoofDiffusion on sparsity-only BuildingNet s99 i0 | height map completion · BuildingNet s99 i0 | RMSE (m): 0.540 (IDW) vs 0.550 | No clear effect |
| use No-FP RoofDiffusion vs CU-Net on PoznanRD s99 i60 and BuildingNet s99 i0 | no-footprint completion · RMSE on PoznanRD s99 i60 and BuildingNet s99 i0 | RMSE (m): 2.600 vs 2.554; 0.962 vs 0.829 | Mixed |
| preprocess sparse point clouds with RoofDiffusion before City3D | City3D · PoznanRD s99 i30 | RMSE (m): 0.352 -> 0.244 | Improved |
| add tree augmentation in training (ablation) | RoofDiffusion · PoznanRD, 3 trees | RMSE (m): 0.523 -> 0.356 | Improved |
| add tree augmentation in training when there are no trees (ablation) | RoofDiffusion · PoznanRD, 0 trees | RMSE (m): 0.203 -> 0.208 | No clear effect |
| increase inference steps from 60 to 500 (ablation) | RoofDiffusion · PoznanRD s99 i80 | RMSE (m): 1.123 -> 0.893 | Improved |
| increase inference steps beyond 500 to 2000 (ablation) | RoofDiffusion · PoznanRD s99 i80 | RMSE (m): 0.893 -> 0.922 | No clear effect |
| apply RoofDiffusion to KITTI depth completion | RoofDiffusion · KITTI depth completion val | RMSE: 917.76 (CU-Net) vs 1631.98 | Hurt |
| apply mean shift offset to KITTI predictions (ablation) | RoofDiffusion on KITTI · KITTI depth completion val | MAE: 491.61 -> 448.97 | Improved |
Closest papers by meaning
- Building3D: A Urban-Scale Dataset and Benchmarks for Learning Roof Structures from Point Clouds (ICCV 2023 · similarity 0.86)
- Scaling Diffusion Models to Real-World 3D LiDAR Scene Completion (CVPR 2024 · similarity 0.84)
- D$^2$GS: Dense Depth Regularization for LiDAR-free Urban Scene Reconstruction (NeurIPS 2025 · similarity 0.83)
- BWFormer: Building Wireframe Reconstruction from Airborne LiDAR Point Cloud with Transformer (CVPR 2025 · similarity 0.83)
- ComPC: Completing a 3D Point Cloud with 2D Diffusion Priors (ICLR 2025 · similarity 0.83)
- PBWR: Parametric-Building-Wireframe Reconstruction from Aerial LiDAR Point Clouds (CVPR 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
