Research map / Image generation / Fast diffusion sampling
AlignDiff: Aligning Diffusion Models for General Few-Shot Segmentation
Ri-Zhao Qiu, Yu-Xiong Wang, Kris Hauser
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 naive text-conditioned GD synthetic samples to the support set | HSNet on FSS-1000 · FSS-1000 1-shot | overall IoU: 86.5 -> 81.4 | Hurt |
| add semi-supervised mask generation to text-conditioned synthesis (ablation) | HSNet + text-conditioned samples · FSS-1000 1-shot | overall IoU: 81.4 -> 87.2 | Improved |
| add normalized masked textual inversion on top of semi-supervised masks (ablation) | HSNet + AlignDiff · FSS-1000 1-shot (OOD classes) | OOD IoU: 86.9 -> 88.2 | Improved |
| add normalized masked textual inversion on top of semi-supervised masks (ablation) | VAT + AlignDiff · FSS-1000 1-shot (OOD classes) | OOD IoU: 88.6 -> 90.6 | Improved |
| add text-only samples with semi-supervised masks (no inversion) (ablation) | VAT · FSS-1000 1-shot | overall IoU: 90.0 -> 89.8 | No clear effect |
| add AlignDiff synthetic samples to SCCAN | SCCAN · Pascal-5i 1-shot | mean IoU: 67.7 -> 69.0 | Improved |
| add AlignDiff synthetic samples to SCCAN | SCCAN · COCO-20i 5-shot | mean IoU: 55.1 -> 55.4 | Improved |
| add AlignDiff synthetic samples to GAPS | GAPS · Pascal-5i GFSS 1-shot | novel IoU: 23.6 -> 43.3 | Improved |
| replace GD samples with AlignDiff samples | GAPS · Pascal-5i GFSS 1-shot | novel IoU: 41.2 -> 43.3 | Improved |
| add AlignDiff samples to GAPS 5-shot | GAPS · COCO-20i GFSS 5-shot | base IoU: 49.1 -> 47.9 | Mixed |
| add AlignDiff samples to simple fine-tuning | FineTune · Pascal-5i GFSS 1-shot | novel IoU: 3.9 -> 44.9 | Improved |
| combine FSS-model masks with attention-map consensus (AlignDiff) instead of attention maps alone (ablation) | mask generation · Pascal-5^2 reference-labeled, all images | mask IoU: 78.6 -> 92.9 | Improved |
Closest papers by meaning
- Unleashing the Potential of the Diffusion Model in Few-shot Semantic Segmentation (NeurIPS 2024 · similarity 0.89)
- $\mathbf{D^{3}S^{2}}$: Diffusion-Guided Dataset Distillation for Semantic Segmentation (NeurIPS 2026 · similarity 0.88)
- High-Precision Dichotomous Image Segmentation via Probing Diffusion Capacity (ICLR 2025 · similarity 0.87)
- MaskDiff: Modeling Mask Distribution with Diffusion Probabilistic Model for Few-Shot Instance Segmentation (AAAI 2024 · similarity 0.87)
- Seg4Diff: Unveiling Open-Vocabulary Semantic Segmentation in Text-to-Image Diffusion Transformers (NeurIPS 2025 · similarity 0.87)
- Adversarial Supervision Makes Layout-to-Image Diffusion Models Thrive (ICLR 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.
