Research map / Architectures and efficiency / Pruning and efficient tuning
From Holistic to Localized: Local Enhanced Adapters for Efficient Visual Instruction Fine-Tuning
Pengkun Jiao, Bin Zhu, Jingjing Chen, Chong-Wah Ngo, Yu-Gang Jiang
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 Dual-LoRA (ablation) | vanilla LoRA with VCE off · UniFood Ingredient Recognition | IoU: 23.2 -> 24.2 | Improved |
| add Dual-LoRA (ablation) | vanilla LoRA · UniFood Recipe Generation | SacreBLEU: 12.4 -> 14.8 | Improved |
| add VCE to vanilla LoRA (ablation) | vanilla LoRA · UniFood Ingredient Recognition | IoU: 23.2 -> 23.3 | No clear effect |
| add VCE to Dual-LoRA (ablation) | Dual-LoRA · UniFood Ingredient Recognition | IoU: 24.2 -> 24.5 | Improved |
| add VCE to Dual-LoRA (ablation) | Dual-LoRA · UniFood Recipe Generation | SacreBLEU: 14.8 -> 14.7 | No clear effect |
| add skill-space layer normalization (ablation) | Dual-LoRA (frozen projector) · UniFood Ingredient Recognition | IoU: 21.4 -> 22.6 | Improved |
| add task-space non-linear activation (ablation) | Dual-LoRA (frozen projector) · UniFood Ingredient Recognition | IoU: 21.4 -> 21.9 | Improved |
| add both norm and non-linearity (ablation) | Dual-LoRA (frozen projector) · UniFood Ingredient Recognition | IoU: 21.4 -> 23.0 | Improved |
| make vision projector trainable (ablation) | Dual-LoRA · UniFood Ingredient Recognition | IoU: 23.0 -> 24.2 | Improved |
| use Dual-LoRA instead of RoDE | LoRA-MoE comparison · ScienceQA | accuracy: 78.39 -> 79.17 | Improved |
| use Dual-LoRA instead of vanilla LoRA | LoRA · ScienceQA | accuracy: 70.01 -> 79.17 | Improved |
| use Dual-LoRA instead of LoRA-MoE (top-2) | LoRA-MoE · UniFood Nutrition Estimation (avg) | pMAE: 50.1 -> 49.9 | Improved |
Closest papers by meaning
- Hawaii: Hierarchical Visual Knowledge Transfer for Efficient Vision-Language Models (NeurIPS 2025 · similarity 0.91)
- LoRASculpt: Sculpting LoRA for Harmonizing General and Specialized Knowledge in Multimodal Large Language Models (CVPR 2025 · similarity 0.91)
- Adapt-$\infty$: Scalable Continual Multimodal Instruction Tuning via Dynamic Data Selection (ICLR 2025 · similarity 0.90)
- Learning to Instruct for Visual Instruction Tuning (NeurIPS 2025 · similarity 0.89)
- Skip Tuning: Pre-trained Vision-Language Models are Effective and Efficient Adapters Themselves (CVPR 2025 · similarity 0.89)
- VIGC: Visual Instruction Generation and Correction (AAAI 2024 · similarity 0.89)
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.
