Research map / Image generation / Fast diffusion sampling
DatasetDM: Synthesizing Data with Perception Annotations Using Diffusion Models
Weijia Wu, Yuzhong Zhao, Hao Chen, Yuchao Gu, Rui Zhao, Yefei He, Hong Zhou, Mike Zheng Shou, Chunhua Shen
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 40k DatasetDM synthetic images to 100 real images | Mask2Former Swin-B · VOC 2012 | mIoU: 65.2 -> 78.5 | Improved |
| add synthetic data to full real training set | Mask2Former Swin-B · VOC 2012 full | mIoU: 84.3 -> 85.4 | Improved |
| add synthetic data to 800 real images | Mask2Former Swin-B · COCO instance seg | AP: 14.4 -> 26.5 | Improved |
| add synthetic data to 50 real images | Depthformer · NYU Depth V2 | REL: 0.31 -> 0.21 | Improved |
| synthetic long-tail data for tail classes | Mask2Former · VOC long-tail | mIoU: 52.6 -> 70.0 | Improved |
| synthetic only vs DiffuMask on unseen classes | DatasetDM · VOC zero-shot | unseen mIoU: 65.0 -> 60.5 | Hurt |
| use larger diffusion step 800 instead of step 1 (ablation) | DatasetDM · VOC 2012 | mIoU: 78.5 -> 76.1 | Hurt |
| add cross-attention maps at step 1 (ablation) | DatasetDM · VOC 2012 | mIoU: 77.6 -> 78.5 | Improved |
| increase training images 60 -> 1,000 (ablation) | DatasetDM · VOC 2012 | mIoU (joint): 77.1 -> 81.1 | Improved |
| GPT-4 prompts instead of human prompts (ablation) | DatasetDM · VOC 2012 | mIoU: 76.6 -> 77.1 | Improved |
| increase GPT-4 prompt candidates 100 -> 1k (ablation) | DatasetDM · VOC 2012 | mIoU: 77.1 -> 78.5 | Improved |
| increase synthetic images 100 -> 40k (ablation) | DatasetDM joint · VOC 2012 | mIoU: 67.9 -> 78.5 | Improved |
Later papers that compared with this method
3 later accepted papers report DatasetDM as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- MMDiff: Extending Diffusion Transformers for Multi-Modal Generation (ECCV 2026 · better on 3 of 3 shared results)
- FreeGen: Bridging Visual-Linguistic Discrepancies Towards Diffusion-based Pixel-level Data Synthesis (AAAI 2025 · better on 2 of 2 shared results)
- Open-Vocabulary Attention Maps with Token Optimization for Semantic Segmentation in Diffusion Models (CVPR 2024 · better on 2 of 2 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
- $\mathbf{D^{3}S^{2}}$: Diffusion-Guided Dataset Distillation for Semantic Segmentation (NeurIPS 2026 · similarity 0.88)
- DEEM: Diffusion models serve as the eyes of large language models for image perception (ICLR 2025 · similarity 0.88)
- Diff-2-in-1: Bridging Generation and Dense Perception with Diffusion Models (ICLR 2025 · similarity 0.87)
- Scaling Properties of Diffusion Models For Perceptual Tasks (CVPR 2025 · similarity 0.87)
- Aligning Generative Denoising with Discriminative Objectives Unleashes Diffusion for Visual Perception (ICLR 2025 · similarity 0.87)
- What Matters When Repurposing Diffusion Models for General Dense Perception Tasks? (ICLR 2025 · similarity 0.86)
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.
