Research map / Vision-language and multimodal / CLIP and zero-shot
AMCI: Unlock the Potential of Large Multimodal Models for Fine-grained Open-world Classification via Adaptive Memory Context Injection
Xiao Liu, Haiyang Zheng, Nan Pu, Wenjing Li, Nicu Sebe, Zhun Zhong
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 AMCI memory context injection | Qwen2VL 7B · Fine-grained (5 datasets avg) | TI: 21.0 -> 41.5 | Improved |
| add AMCI | LLaVA-OV 0.5B · Fine-grained (5 datasets avg) | TI: 3.9 -> 13.8 | Improved |
| add AMCI | Qwen2VL 7B · Low-resource (HAM10000, iNaturalist avg) | TI: 1.8 -> 5.2 | Improved |
| add memory module (ablation) | AMCI · Oxford Pets, FGVC Aircraft, HAM10000 | avg across metrics: 12.0% | Improved |
| add LLM refinement (ablation) | AMCI · Oxford Pets, FGVC Aircraft, HAM10000 | avg across metrics: 25.2% | Improved |
| use EMA dynamic threshold instead of FIFO replacement (ablation) | AMCI memory · Oxford Pets, FGVC Aircraft, HAM10000 | avg across metrics: 4.1% | Improved |
| use EMA dynamic threshold instead of LRU replacement (ablation) | AMCI memory · Oxford Pets, FGVC Aircraft, HAM10000 | avg across metrics: 4.0% | Improved |
| use EMA dynamic threshold instead of fixed threshold (ablation) | AMCI memory · Oxford Pets, FGVC Aircraft, HAM10000 | avg across metrics: 1.2% | Improved |
| use similarity retrieval instead of random retrieval (ablation) | AMCI · Oxford Pets, FGVC Aircraft, HAM10000 | avg across metrics: 8.2% | Improved |
| retrieve k neighbors instead of top-1 (ablation) | AMCI · Oxford Pets, FGVC Aircraft, HAM10000 | avg across metrics: 4.5% | Improved |
| retrieve with visual features instead of CLIP textual features (ablation) | AMCI · Oxford Pets, FGVC Aircraft, HAM10000 | avg across metrics: 1.2% | Improved |
| use attribute-rich prompt instead of minimal prompt (ablation) | AMCI · Oxford Pets, FGVC Aircraft, HAM10000 | avg across metrics: 9.5% | Improved |
Closest papers by meaning
- ContextNav: Towards Agentic Multimodal In-Context Learning (ICLR 2026 · similarity 0.87)
- Specificity-aware reinforcement learning for fine-grained open-world classification (CVPR 2026 · similarity 0.86)
- INQUIRE: A Natural World Text-to-Image Retrieval Benchmark (NeurIPS 2024 · similarity 0.85)
- On the Out-Of-Distribution Generalization of Large Multimodal Models (CVPR 2025 · similarity 0.85)
- Ranked from Within: Ranking Large Multimodal Models Without Labels (ICML 2025 · similarity 0.84)
- Zooming without Zooming: Region-to-Image Distillation for Fine-Grained Multimodal Perception (ICML 2026 · similarity 0.84)
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.
