Research map / Detection and segmentation / Weakly supervised detection and segmentation
Neural Diffusion Distance for Image Segmentation
Jian Sun, Zongben Xu
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 |
|---|---|---|---|
| share classifier between classification and segmentation (ablation) | weakly supervised segmentation network · PASCAL VOC 2012 val | mIoU: 64.7 -> 65.8 | Improved |
| use diffusion distance instead of embedding similarity (ablation) | weakly supervised segmentation network · PASCAL VOC 2012 val | mIoU: 54.7 -> 65.8 | Improved |
| replace global average pooling with PC-WAP and RFP | weakly supervised segmentation network · PASCAL VOC 2012 val | mIoU: 45.2 -> 65.8 | Improved |
| add feature-attentional interpolation of segmentation masks (HR) (ablation) | hierarchical segmentation · BSD500 test | covering MAX: 0.78 -> 0.80 | Improved |
| add feature-attentional interpolation of segmentation masks (HR) (ablation) | hierarchical segmentation · BSD500 test | covering AVR: 0.68 -> 0.69 | Improved |
| replace DeepNCut with neural diffusion distance (HR) | hierarchical segmentation · BSD500 test | covering MAX: 0.70 -> 0.80 | Improved |
| replace NCut-DF with neural diffusion distance (HR) | hierarchical segmentation · BSD500 test | covering AVR: 0.48 -> 0.69 | Improved |
| increase iterations T from 1 to 2 at t=10 (ablation) | spec-diff-net · BSD500 test | cosine similarity: 0.709 -> 0.741 | Improved |
| increase initial t from 5 to 10 at T=2 (ablation) | spec-diff-net · BSD500 test | cosine similarity: 0.738 -> 0.741 | Improved |
| increase initial t from 5 to 10 at T=1 (ablation) | spec-diff-net · BSD500 train+val | cosine similarity: 0.778 -> 0.785 | Improved |
| increase iterations T from 2 to 3 at t=5 (ablation) | spec-diff-net · BSD500 train+val | cosine similarity: 0.785 -> 0.777 | No clear effect |
| increase iterations T from 2 to 3 at t=10 (ablation) | spec-diff-net · BSD500 train+val | cosine similarity: 0.794 -> 0.785 | Hurt |
Closest papers by meaning
- Unleashing the Potential of the Diffusion Model in Few-shot Semantic Segmentation (NeurIPS 2024 · similarity 0.86)
- $\mathbf{D^{3}S^{2}}$: Diffusion-Guided Dataset Distillation for Semantic Segmentation (NeurIPS 2026 · similarity 0.85)
- EmerDiff: Emerging Pixel-level Semantic Knowledge in Diffusion Models (ICLR 2024 · similarity 0.85)
- High-Precision Dichotomous Image Segmentation via Probing Diffusion Capacity (ICLR 2025 · similarity 0.85)
- DiffCut: Catalyzing Zero-Shot Semantic Segmentation with Diffusion Features and Recursive Normalized Cut (NeurIPS 2024 · similarity 0.85)
- Explore In-Context Segmentation via Latent Diffusion Models (AAAI 2025 · similarity 0.85)
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.
