AI Paper Atlas

Research map / Vision-language and multimodal / CLIP and zero-shot

Unlearning without Forgetting: Securely Removing Targeted Concepts from Large-Scale Vision-Language Open-Vocabulary Detectors

Zhongze Wu, Xiu Su, Feng Yang, Dan Niu, Shan You, Yueyi Luo, Jun Long

CVPR 2026 · Vision-language and multimodal › CLIP and zero-shot · approach: Open-vocabulary alignment from VLMs

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 changedWhereMeasuredEffect
constrain updates to the null-space of retain embeddings and add one-step mean-flow forgetting on top of null-space only (ablation)SafeDetect · UOD-Bench OD 1% forget, LLM-Det Swin-TU-Score: 16.7 -> 22.2Improved
add cross-modal decoupling loss on decoder queries (ablation)SafeDetect · UOD-Bench OD 1% forget, LLM-Det Swin-TU-Score: 22.2 -> 23.5Improved
add cross-modal decoupling loss on decoder queries (ablation)SafeDetect · UOD-Bench OD 10% forget, LLM-Det Swin-TU-Score: 8.6 -> 9.7Improved
add null-space projection to preserve zero-shot generalization (ablation)SafeDetect · LVIS-minival, average AP dropLVIS AP drop: 9.6 -> 3.8Improved
apply decoupling to decoder query features rather than bbox head features (ablation)SafeDetect · UOD-Bench OD 1% forget, LLM-Det Swin-TU-Score: 19.8 -> 23.5Improved
decouple on bbox head features instead of no decoupling (ablation)SafeDetect · UOD-Bench OD 1% forget, LLM-Det Swin-TU-Score: 22.2 -> 19.8Hurt
use SafeDetect instead of NPO for unlearningLLM-Det Swin-T · UOD-Bench OD 1% forget, forget mAPforget mAP@50: 50.5 -> 17.8Improved
use SafeDetect instead of NPO, test with synonym class namesLLM-Det Swin-T · UOD-Bench OD 1% forget, forget mAP with synonymsforget mAP@50: 65.8 -> 20.7Improved
use SafeDetect instead of NPOLLM-Det Swin-T · LVIS AP after unlearning, average dropLVIS AP drop: 14.2 -> 8.3Improved
use gradient ascent (GA) for unlearningLLM-Det Swin-T · UOD-Bench OD 1% forgetU-Score: 58.0 -> 57.2 forget mAP; U-Score 2.7No clear effect
use SafeDetect instead of NPO for training convergenceunlearning training · UOD-Bench training losssteps to stabilize: 750 -> 500 stepsImproved
use SafeDetect instead of NPO for compute costunlearning training · UOD-BenchGPU-hours: 22.0 -> 13.2Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy