Research map / Continual learning / Class-incremental
Forget What Has Seen: Selective Concept Unlearning in Segmentation Foundation Models
Miaozeng Du, Jiaqi Li, Sirui Pan, Yi Zhan, Guilin Qi, Yuxin Zhang, Rihui Jin, Yinjia Shu, Qianshan Wei
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 negative distillation level (ablation) | SCU · SAM1 concept unlearning (8 concepts) | Efficacy IoU (lower better): 32.8 -> 16.3 | Improved |
| add attention-level suppression (ablation) | SCU · SAM1 concept unlearning (8 concepts) | Efficacy IoU (lower better): 26.5 -> 16.3 | Improved |
| add output-level relabeling (ablation) | SCU · SAM1 concept unlearning (8 concepts) | Efficacy IoU (lower better): 27.9 -> 16.3 | Improved |
| add output-level relabeling (ablation) | SCU · SAM1 concept unlearning (8 concepts) | Specificity IoU: 32.5 -> 61.3 | Improved |
| add Preservation Module (ablation) | SCU · SAM1 concept unlearning (8 concepts) | Specificity IoU: 21.4 -> 61.3 | Improved |
| add Preservation Module (ablation) | SCU · SAM1 concept unlearning (8 concepts) | Efficacy IoU (lower better): 17.6 -> 16.3 | No clear effect |
| use SCU instead of Gradient Ascent | SCU · SAM1 concept unlearning | Specificity IoU: 16.1 -> 61.3 | Improved |
| use SCU instead of Boundary Shrink / Random Labeling / Boundary Expand | SCU · SAM1 concept unlearning | Efficacy IoU (lower better): 63.4 -> 16.3 (vs RL) | Improved |
| use SCU instead of L1-sparse | SCU · SAM1 concept unlearning | Normalized Score: 0.723 -> 0.923 | Improved |
| use SCU on SAM2 instead of SAM1 | SCU · SCU, SAM1 vs SAM2 | Normalized Score: 0.923 -> 0.929 | No clear effect |
| increase unlearning training steps | SCU · SAM1 steps sweep | Specificity: SCU stable; GA and RL drop | Improved |
Closest papers by meaning
- Targeted Forgetting of Image Subgroups in CLIP Models (CVPR 2025 · similarity 0.88)
- Wisdom is Knowing What not to Say: Hallucination-Free LLMs Unlearning via Attention Shifting (NeurIPS 2025 · similarity 0.88)
- Single Image Unlearning: Efficient Machine Unlearning in Multimodal Large Language Models (NeurIPS 2024 · similarity 0.87)
- Selective Forgetting: Advancing Machine Unlearning Techniques and Evaluation in Language Models (AAAI 2025 · similarity 0.87)
- Decoupled Distillation to Erase: A General Unlearning Method for Any Class-centric Tasks (CVPR 2025 · similarity 0.87)
- Learning What to Forget: Improving LLM Unlearning via Learned Token-Level Importance (NeurIPS 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.
