Research map / Vision-language and multimodal / CLIP and zero-shot
Injection Without Distortion: Geometrically Constrained Knowledge Enhancement for Vision-Language Models
Zhongze Wu, Xiu Su, Feng Yang, Shan You, Jun Long, Yueyi Luo
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 null-space projection to knowledge injection (ablation) | HKI knowledge injection · iNatLoc average (Swin-B, IV) | mAP50: 76.8 -> 79.9 | Improved |
| add null-space projection (ablation) | HKI knowledge injection · iNatLoc L1 (Swin-B, IV) | mAP50: 62.1 -> 67.5 | Improved |
| unconstrained injection (SHiNe) on coarse concepts (ablation) | Detic · iNatLoc L1 (Swin-B, IV) | mAP50: 64.4 -> 62.1 | Hurt |
| protect query concept plus its sub-concepts (hierarchical) vs query-only (ablation) | GeCoin · iNatLoc average (Swin-B, IV) | mAP50: 78.7 -> 79.9 | Improved |
| use fine-grained knowledge (FKI) with projection vs unconstrained (ablation) | FKI knowledge injection · iNatLoc average (Swin-B, IV) | mAP50: 76.6 -> 79.5 | Improved |
| add 20% irrelevant concepts to LLM knowledge (ablation) | GeCoin vs SHiNe · iNatLoc average (ResNet-50), GPT-3.5 | mAP50: GeCoin 51.4 -> 49.5; SHiNe 46.5 -> 43.0 | Hurt |
| GeCoin vs SHiNe on ImageNet BREEDS L1 | GeCoin · ImageNet-1k BREEDS L1 | top-1 accuracy: 50.4 -> 61.3 | Improved |
| GeCoin vs H-CLIP on BREEDS average | GeCoin · ImageNet-1k BREEDS average | top-1 accuracy: 68.4 -> 68.6 | No clear effect |
| GeCoin vs SHiNe on FSOD L3 | GeCoin · FSOD (Swin-B, IV) | mAP50: 66.4 -> 66.5 | No clear effect |
| GeCoin on CoDet | CoDet · iNatLoc average (ResNet-50) | mAP50: 34.4 -> 38.6 (vs SHiNe) | Improved |
| GeCoin on LLM-Det | LLM-Det · iNatLoc average (Swin-T) | mAP50: 56.7 -> 59.1 (vs SHiNe) | Improved |
Closest papers by meaning
- Cyclic Contrastive Knowledge Transfer for Open-Vocabulary Object Detection (ICLR 2025 · similarity 0.88)
- Synthesize Diagnose and Optimize: Towards Fine-Grained Vision-Language Understanding (CVPR 2024 · similarity 0.88)
- Seeing Clearly, Reasoning Confidently: Plug-and-Play Remedies for Vision Language Model Blindness (CVPR 2026 · similarity 0.88)
- The Neglected Tails in Vision-Language Models (CVPR 2024 · similarity 0.88)
- VLKEB: A Large Vision-Language Model Knowledge Editing Benchmark (NeurIPS 2024 · similarity 0.87)
- Large Vision-Language Models Get Lost in Attention (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.
