Research map / Vision-language and multimodal / Multimodal LLMs
Beyond LLaVA-HD: Diving Into High-Resolution Multimodal Large Language Models
Yifan Zhang, Qingsong Wen, Chaoyou Fu, Kun Wang, Xue Wang, Zhang Zhang, Liang Wang, Rong Jin
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 mixture of global adapters (ablation) | SliME baseline · ScienceQA | accuracy: 74.30 -> 77.00 | Improved |
| add mixture of global adapters (ablation) | SliME baseline · TextVQA | accuracy: 58.94 -> 60.57 | Improved |
| local compression with 144 queries vs 64 (ablation) | SliME · TextVQA | accuracy: 63.09 -> 63.58 | Improved |
| increase local queries from 144 to 196 (ablation) | SliME · TextVQA | accuracy: 63.58 -> 62.48 | Hurt |
| router keeps 90% of local tokens instead of all (ablation) | SliME · TextVQA | accuracy: 63.58 -> 64.05 | Improved |
| router keeps 50% of local tokens (ablation) | SliME · TextVQA | accuracy: 64.05 -> 62.75 | Hurt |
| router keeps 50% instead of 90% (ablation) | SliME · POPE | accuracy: 84.20 -> 84.90 | Improved |
| add SMR reasoning data (ablation) | SliME · ScienceQA | accuracy: 77.09 -> 82.29 | Improved |
| add SMR reasoning data (ablation) | SliME · TextVQA | accuracy: 63.58 -> 59.89 | Hurt |
| alternating instead of end-to-end training (ablation) | SliME · TextVQA | accuracy: 59.69 -> 63.58 | Improved |
| alternating instead of end-to-end training (ablation) | SliME · ScienceQA | accuracy: 74.12 -> 77.44 | Improved |
| alternating instead of end-to-end training (ablation) | SliME · AMBER | accuracy: 86.00 -> 88.40 | Improved |
Closest papers by meaning
- Feast Your Eyes: Mixture-of-Resolution Adaptation for Multimodal Large Language Models (ICLR 2025 · similarity 0.90)
- Divide, Conquer and Combine: A Training-Free Framework for High-Resolution Image Perception in Multimodal Large Language Models (AAAI 2025 · similarity 0.90)
- iLLaVA: An Image is Worth Fewer Than 1/3 Input Tokens in Large Multimodal Models (ICLR 2026 · similarity 0.90)
- Matryoshka Multimodal Models (ICLR 2025 · similarity 0.90)
- LLaVA-Mini: Efficient Image and Video Large Multimodal Models with One Vision Token (ICLR 2025 · similarity 0.90)
- HiRes-LLaVA: Restoring Fragmentation Input in High-Resolution Large Vision-Language Models (CVPR 2025 · 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.
