Research map / Image generation / Autoregressive and latent generation
LayoutDM: Discrete Diffusion Model for Controllable Layout Generation
Naoto Inoue, Kotaro Kikuchi, Edgar Simo-Serra, Mayu Otani, Kota Yamaguchi
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 discrete diffusion with modality-wise corruption, decoupled positional encoding, and adaptive quantization (ablation) | LayoutDM · Rico, unconditional | FID: 6.65 -> 7.32 | Improved |
| Add modality-wise diffusion (ablation) | LayoutDM · Rico, unconditional | Alignment: 0.156 -> 0.162 | Hurt |
| Add decoupled positional encoding (ablation) | LayoutDM · Rico, unconditional | Alignment: 0.227 -> 0.162 | Improved |
| Use KMeans quantization instead of uniform quantization (ablation) | LayoutDM · Rico, unconditional | FID: 7.58 -> 6.65 | Improved |
| Use KMeans quantization instead of percentile quantization (ablation) | LayoutDM · Rico, unconditional | FID: 9.79 -> 6.65 | Improved |
| Use KMeans quantization instead of uniform quantization (ablation) | LayoutDM · Rico, unconditional | Alignment: 0.256 -> 0.162 | Improved |
| Use KMeans quantization instead of percentile quantization (ablation) | LayoutDM · Rico, unconditional | Alignment: 0.232 -> 0.162 | Improved |
| Add weak geometric logit adjustment to masking-based refinement (ablation) | LayoutDM · Rico, refinement | FID: 3.55 -> 2.77 | Improved |
| Use the default prior instead of a Gaussian prior (ablation) | LayoutDM · Rico, refinement | FID: 5.82 -> 2.77 | Improved |
| Use the default prior instead of a negation prior (ablation) | LayoutDM · Rico, refinement | FID: 3.78 -> 2.77 | Improved |
| Use discrete rather than continuous state-space diffusion (ablation) | LayoutDM versus DiffusionLM · Rico, unconditional | FID: 24.8 -> 6.65 | Improved |
| Use LayoutDM rather than the strongest baseline | LayoutDM · Rico, C→S+P | FID: 3.97 -> 3.55 | Improved |
Later papers that compared with this method
3 later accepted papers report LayoutDM as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- Towards Aligned Layout Generation via Diffusion Model with Aesthetic Constraints (ICLR 2024 · better on 14 of 16 shared results)
- Visual Layout Composer: Image-Vector Dual Diffusion Model for Design Layout Generation (CVPR 2024 · better on 7 of 8 shared results)
- LayoutFlow: Flow Matching for Layout Generation (ECCV 2024 · better on 4 of 4 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- LayoutDiffusion: Improving Graphic Layout Generation by Discrete Diffusion Probabilistic Models (ICCV 2023 · similarity 0.92)
- Layout-Corrector: Alleviating Layout Sticking Phenomenon in Discrete Diffusion Model (ECCV 2024 · similarity 0.92)
- Towards Aligned Layout Generation via Diffusion Model with Aesthetic Constraints (ICLR 2024 · similarity 0.91)
- DLT: Conditioned layout generation with Joint Discrete-Continuous Diffusion Layout Transformer (ICCV 2023 · similarity 0.90)
- Visual Layout Composer: Image-Vector Dual Diffusion Model for Design Layout Generation (CVPR 2024 · similarity 0.88)
- MaskPLAN: Masked Generative Layout Planning from Partial Input (CVPR 2024 · similarity 0.87)
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.
