Research map / Image generation / Fast diffusion sampling
Diffusion Model as a Generalized Segmentation Learner
Haoxiao Wang, Antao Xiang, Haiyang Sun, Peilin Sun, Changhao Pan, Yifu Chen, Minjie Hong, Weijie Wang, Shuang Chen, Yue Chen, ZHOU ZHAO
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 |
|---|---|---|---|
| add multi-resolution noise (ablation) | DiGSeg · COCO | mIoU: 48.9 -> 49.7 | Improved |
| add annealed noise alone (ablation) | DiGSeg · COCO | mIoU: 48.9 -> 49.2 | Improved |
| combine multi-resolution and annealed noise (ablation) | DiGSeg · ADE20K | mIoU: 56.7 -> 58.6 | Improved |
| ensemble 8 predictions with 2 DDIM-trailing steps instead of 1x1 (ablation) | DiGSeg · ADE20K | mIoU: 56.8 -> 58.6 | Improved |
| ensemble 8 predictions instead of 1 prediction (ablation) | DiGSeg · COCO | mIoU: 48.2 -> 50.8 | Improved |
| 20 ensembles with 50 steps without trailing instead of 8x2 (ablation) | DiGSeg · ADE20K | mIoU: 58.6 -> 58.8 | No clear effect |
| use DDIM-trailing timesteps (ablation) | DiGSeg · inference speed | FPS: 0.12 -> 3.15 | Improved |
| train on ADE20K only vs COCO only, test on Cityscapes (ablation) | DiGSeg · Cityscapes | mIoU: 38.51 -> 43.58 | Improved |
| train on COCO+ADE20K vs ADE20K only, test on Cityscapes (ablation) | DiGSeg · Cityscapes | mIoU: 43.58 -> 39.04 | Hurt |
| train on half of ADE20K (ablation) | DiGSeg · ADE20K | mIoU: not reported numerically | No clear effect |
| apply DiGSeg to medical REFUGE-2 without domain design | DiGSeg · REFUGE-2 | IoU: 53.5 -> 34.5 | Hurt |
| DiGSeg vs EoMT on COCO semantic segmentation | DiGSeg · COCO | mIoU: 48.7 -> 50.8 | Improved |
Closest papers by meaning
- Unleashing the Potential of the Diffusion Model in Few-shot Semantic Segmentation (NeurIPS 2024 · similarity 0.91)
- Seg4Diff: Unveiling Open-Vocabulary Semantic Segmentation in Text-to-Image Diffusion Transformers (NeurIPS 2025 · similarity 0.91)
- EmerDiff: Emerging Pixel-level Semantic Knowledge in Diffusion Models (ICLR 2024 · similarity 0.91)
- VidSeg: Training-free Video Semantic Segmentation based on Diffusion Models (CVPR 2025 · similarity 0.90)
- Factorized Diffusion Architectures for Unsupervised Image Generation and Segmentation (NeurIPS 2024 · similarity 0.90)
- Prompting Diffusion Models for Zero-Shot Instance Segmentation (NeurIPS 2026 · similarity 0.90)
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.
