Research map / Graph learning / Graph self-supervised learning
COHO: Context-Sensitive City-Scale Hierarchical Urban Layout Generation
Liu He, Daniel Aliaga
Read the paperCheck your idea against 137,217 papers
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 GAT backbone in block VAE instead of GCN or SAGE (ablation) | BVAE · Block-level reconstruction | Overlap (%): 0.45 -> 0.17 | Improved |
| use per-dimension quantization (L=20) instead of trainable VQ (ablation) | BVAE latent quantization · Block quantization | Geom-E (%): 5.78 -> 0.89 | Improved |
| use per-dimension quantization instead of KMeans clustering (ablation) | BVAE latent quantization · Block quantization | Geom-E (%): 17.85 -> 0.89 | Improved |
| increase quantization levels per dimension from 20 to 30 (ablation) | DIM-Q · Block quantization | Geom-E (%): 0.89 -> 0.78 | No clear effect |
| use GAT message passing instead of GCN in city-level GMAE (ablation) | GMAE · City-scale layout | FID: 52.95 -> 23.63 | Improved |
| increase encoder depth from 1 to 3 hops (ablation) | GMAE · City-scale layout | FID: 43.75 -> 23.63 | Improved |
| increase encoder depth from 3 to 4 (ablation) | GMAE · City-scale layout | FID: 23.63 -> 24.59 | No clear effect |
| use GAT decoder with remasking instead of MLP decoder (ablation) | GMAE · City-scale layout | FID: 23.63 -> 48.24 | Hurt |
| use dynamic large mask ratio instead of fixed 0.15 (ablation) | GMAE training · City-scale layout | FID: 49.50 -> 23.63 | Improved |
| use cosine scheduling instead of linear or logarithmic (ablation) | GMAE iterative generation · City-scale layout | FID: 37.42 -> 23.63 (linear), 63.25 (log) | Improved |
| use more iterations T=12 instead of single pass (ablation) | GMAE iterative generation · City-scale layout | FID: 76.12 -> 23.63 | Improved |
| COHO vs baselines on overlap metric | COHO · US city-scale layout test | Overlap: VTN 1.24 vs COHO 1.27 | Mixed |
Closest papers by meaning
- GlobalMapper: Arbitrary-Shaped Urban Layout Generation (ICCV 2023 · similarity 0.89)
- CityDreamer: Compositional Generative Model of Unbounded 3D Cities (CVPR 2024 · similarity 0.87)
- BlockPlanner: City Block Generation With Vectorized Graph Representation (ICCV 2021 · similarity 0.86)
- Text2City: One-Stage Text-Driven Urban Layout Regeneration (AAAI 2024 · similarity 0.86)
- MaskPLAN: Masked Generative Layout Planning from Partial Input (CVPR 2024 · similarity 0.86)
- Rethinking The Training And Evaluation of Rich-Context Layout-to-Image Generation (NeurIPS 2024 · similarity 0.84)
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.
