Research map / Image generation / Text-to-image diffusion
MapRoute:Precise-Concept Erasing Mappers via Semantic Routing
Sihao Li, Baixi Liang, Shuohong Xia, Yunyun Yang
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 Lkeep1 preservation loss over the concept dictionary (ablation) | MapRoute Mapper · CIFAR-10 object erasure | ACCs: 19.49 -> 99.53 | Improved |
| add Lkeep2 preservation loss over celebrity names (ablation) | MapRoute Mapper · Artistic style erasure | CLIPe: 18.5 -> 16.85 | Improved |
| use three Map layers instead of two (ablation) | Mapper · CIFAR-10 object erasure | ACCs: 16.33 -> 99.53 | Improved |
| use four Map layers instead of three (ablation) | Mapper · CIFAR-10 object erasure | ACCe: 0.5 -> 0.43 | No clear effect |
| use MapRoute instead of MACE | Stable Diffusion v1.4 · CIFAR-10 object erasure (10 classes) | Ho: 92.61 -> 99.37 | Improved |
| use MapRoute instead of MACE for 50-celebrity multi-concept erasure | Stable Diffusion v1.4 · Celebrity erasure (50 celebrities) | Hc: 85.95 -> 94.83 | Improved |
| use MapRoute instead of MACE for style erasure | Stable Diffusion v1.4 · Artistic style erasure (Brent Heighton) | Ha: 6.82 -> 12.21 | Improved |
| use MapRoute instead of MACE for style erasure | Stable Diffusion v1.4 · Artistic style erasure (Brett Weston) | CLIPe: 15.84 -> 17.78 | Hurt |
| choose an unrelated surrogate concept for the target (ablation) | MapRoute Mapper · Celebrity erasure | erasure quality | No clear effect |
| MapRoute instead of MACE in mixed concept erasure | Stable Diffusion v1.4 · Mixed concepts erasure | ACCs: 83.48 -> 98.37 | Improved |
Closest papers by meaning
- EraseAnything: Enabling Concept Erasure in Rectified Flow Transformers (ICML 2025 · similarity 0.90)
- Semantic Surgery: Zero-Shot Concept Erasure in Diffusion Models (NeurIPS 2025 · similarity 0.90)
- Consistency-Preserving Concept Erasure via Unsafe–Safe Pairing and Directional Fisher-weighted Adaptation (NeurIPS 2026 · similarity 0.89)
- MACE: Mass Concept Erasure in Diffusion Models (CVPR 2024 · similarity 0.89)
- Differential Vector Erasure: Unified Training-Free Concept Erasure for Flow Matching Models (NeurIPS 2026 · similarity 0.89)
- Precise, Fast, and Low-cost Concept Erasure in Value Space: Orthogonal Complement Matters (CVPR 2025 · similarity 0.88)
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.
