Research map / Image generation / Fast diffusion sampling
MaskDiME: Adaptive Masked Diffusion for Precise and Efficient Visual Counterfactual Explanations
Changlu Guo, Anders Nymark Christensen, Anders Bjorholm Dahl, Morten Rieger Hannemose
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 |
|---|---|---|---|
| replace recursive clean-image estimate with one-step estimate (s=1, no mask) (ablation) | DiME · CelebA Smile | FID: 3.17 -> 95.76 | Hurt |
| increase gradient scale s from 1 to 8 (ablation) | MaskDiME without mask · CelebA Smile | flip rate (%): 55.2 -> 100.0 | Improved |
| increase gradient scale s from 1 to 8 (ablation) | MaskDiME without mask · CelebA Smile | FID: 95.76 -> 15.94 | Improved |
| add fixed mask (ablation) | s=8 & w/o mask · CelebA Smile | FID: 15.94 -> 4.21 | Improved |
| use adaptive dual-mask (rho=1) instead of fixed mask (ablation) | MaskDiME · CelebA Smile | FID: 4.21 -> 0.71 | Improved |
| shrink clean-level mask with rho=0.5 (ablation) | MaskDiME with rho=1 · CelebA Smile | COUT: 0.81 -> 0.87 | Improved |
| use MaskDiME instead of DiME | DiME · CelebA Smile | runtime: over 30x speedup | Improved |
| use MaskDiME instead of ACE/RCSB | ACE, RCSB · CelebA Smile | peak GPU memory: about one-tenth | Improved |
| use MaskDiME instead of ACE l1 | ACE l1 · BDD100K | FID: 1.02 -> 3.19 | Hurt |
| use MaskDiME instead of RCSB | RCSB · ImageNet Sorrel-Zebra | FID: 8.0 -> 32.5 | Hurt |
| use MaskDiME instead of DiME | DiME · CelebA diversity | LPIPS diversity: 0.2139 -> 0.0395 | Hurt |
| use MaskDiME instead of DiME on CD | DiME · CelebA Smile | CD: 2.30 -> 2.41 | Mixed |
Closest papers by meaning
- Diffusion Counterfactual Generation with Semantic Abduction (ICML 2025 · similarity 0.89)
- V-CECE: Visual Counterfactual Explanations via Conceptual Edits (NeurIPS 2025 · similarity 0.89)
- Rethinking Visual Counterfactual Explanations Through Region Constraint (ICLR 2025 · similarity 0.89)
- Simplified and Generalized Masked Diffusion for Discrete Data (NeurIPS 2024 · similarity 0.87)
- Envision, Attend, Then Respond: Counterfactual Hallucination Mitigation in Large Vision-Language Models (CVPR 2026 · similarity 0.87)
- Mitigating Mask Prior Drift and Positional Attention Collapse in Large Diffusion Vision-Language Models (ICML 2026 · 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.
