Research map / Image generation / Fast diffusion sampling
Stochastic Segmentation with Conditional Categorical Diffusion Models
Lukas Zbinden, Lars Doorenbos, Theodoros Pissas, Adrian Thomas Huber, Raphael Sznitman, Pablo Márquez-Neila
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 |
|---|---|---|---|
| average predictions over more samples for label map fusion (ablation) | CCDM · Cityscapes 128x256 | mIoU: 53.2 -> 56.2 | Improved |
| add Dino ViT-S features as conditioning (ablation) | CCDM · Cityscapes 256x512 (single sample) | mIoU: 60.3 -> 64.0 | Improved |
| add Dino ViT-S features as conditioning (ablation) | CCDM · Cityscapes 128x256 (single sample) | mIoU: 53.2 -> 55.5 | Improved |
| increase CCDM capacity from 5.4M to 30M parameters (ablation) | CCDM · Cityscapes 128x256 | mIoU (samples=1): 37.8 -> 53.2 | Improved |
| increase CCDM size from 9M to 41M (ablation) | CCDM · LIDCv1 | GED_16: 0.212 -> 0.207 | Improved |
| train CCDM categorical diffusion instead of cross-entropy UNet (ablation) | UNet · Cityscapes 128x256 | mIoU: 48.7 -> 53.2 | Improved |
| reduce number of inference steps via step skipping (ablation) | CCDM · LIDC | GED_16, HM-IoU_16 | Hurt |
| fuse samples by averaging probabilities instead of majority vote (ablation) | CCDM · Cityscapes | mIoU | Improved |
| use CCDM instead of MoSE with larger model | LIDCv1 · LIDCv1 | HM-IoU_16: 0.623 vs 0.624 | No clear effect |
Closest papers by meaning
- HSRDiff: A Hierarchical Self-Regulation Diffusion Model for Stochastic Semantic Segmentation (AAAI 2025 · similarity 0.89)
- Stochastic Conditional Diffusion Models for Robust Semantic Image Synthesis (ICML 2024 · similarity 0.89)
- Diffusion-Driven Two-Stage Active Learning for Low-Budget Semantic Segmentation (NeurIPS 2025 · similarity 0.88)
- EmerDiff: Emerging Pixel-level Semantic Knowledge in Diffusion Models (ICLR 2024 · similarity 0.88)
- Unleashing the Potential of the Diffusion Model in Few-shot Semantic Segmentation (NeurIPS 2024 · similarity 0.88)
- $\mathbf{D^{3}S^{2}}$: Diffusion-Guided Dataset Distillation for Semantic Segmentation (NeurIPS 2026 · similarity 0.88)
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.
