Research map / Detection and segmentation / Open-set and few-shot detection
Retain and Adapt: Auto-Balanced Model Editing for Open-Vocabulary Object Detection under Domain Shifts
Zixuan Duan, Fengyuan Lu, Xunzhi Xiang, Wenbin Li, Yang Gao, Qi Fan
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 |
|---|---|---|---|
| edit FFN output layers with auto-balanced KV objective instead of continual learning baselines | Grounding DINO · CDFSOD-6 (5-shot) | AP: 31.0 (EWC) -> 38.5 | Improved |
| use auto-balanced Gamma regularizer instead of manually tuned lambda (ablation) | ABME · CDFSOD-6, 1 to 6 tasks | new-task mAP | Improved |
| use exponent 1/4 instead of 1/2 in Gamma (ablation) | ABME · CDFSOD-6 (5-shot) | AP: 36.6 -> 38.5 | Improved |
| use exponent 1/8 instead of 1/4 in Gamma (ablation) | ABME · CDFSOD-6 (5-shot) | AP: 38.5 -> 38.5 | No clear effect |
| edit all FFNs instead of a sub-module only (ablation) | ABME · CDFSOD-6 (5-shot) | new-task AP: 35.1 (all backbone) -> 38.5 | Improved |
| edit FFN output layers instead of self-attention layers (ablation) | ABME · CDFSOD-6 (5-shot) | AP: 36.3 -> 38.5 | Improved |
| add ABME on top of EWC | EWC · CDFSOD-6 (10-shot) | AP: 31.9 -> 41.7 | Improved |
| add ABME on top of SD-LoRA | SD-LoRA · CDFSOD-6 (10-shot) | AP: 28.4 -> 38.5 | Improved |
| aggregate task KV statistics additively rather than sequential fine-tuning | ABME · CDFSOD-6 task order | AP stability | Improved |
| apply ABME to GLIP and CLIP | ABME · CDFSOD-6 GLIP; CLIP 11 datasets | AP / accuracy: GLIP 19.1 -> 35.2 (5-shot); CLIP 65.3 -> 74.1 | Improved |
| ABME versus continual learning on COCO retention | ABME · CDFSOD-6 5-shot COCO AP | COCO AP: Adam-NSCL 57.8 vs Ours 57.0 | Mixed |
Closest papers by meaning
- DitHub: A Modular Framework for Incremental Open-Vocabulary Object Detection (NeurIPS 2025 · similarity 0.91)
- Cyclic Contrastive Knowledge Transfer for Open-Vocabulary Object Detection (ICLR 2025 · similarity 0.90)
- Exploring Region-Word Alignment in Built-in Detector for Open-Vocabulary Object Detection (CVPR 2024 · similarity 0.89)
- Open-Det: An Efficient Learning Framework for Open-Ended Detection (ICML 2025 · similarity 0.88)
- Learning Background Prompts to Discover Implicit Knowledge for Open Vocabulary Object Detection (CVPR 2024 · similarity 0.88)
- CAKE: Category Aware Knowledge Extraction for Open-Vocabulary Object Detection (AAAI 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.
