Research map / Vision-language and multimodal / Multimodal LLMs
LLaVA-UHD: an LMM Perceiving any Aspect Ratio and High-Resolution Images
Zonghao Guo, Ruyi Xu, Yuan Yao, Junbo Cui, Zanlin Ni, Chunjiang Ge, Tat-Seng Chua, Zhiyuan Liu, Gao Huang
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 |
|---|---|---|---|
| replace padding with adaptive aspect-ratio encoding (no padding) (ablation) | LLaVA-1.5 · TextVQA | accuracy: 61.3 -> 61.8 | Improved |
| use perceiver resampler instead of 2-layer MLP projector (ablation) | LLaVA-UHD · 9 benchmarks (TFLOPs) | TFLOPs: 113.65 -> 14.63 TFLOPs, TextVQA 66.9 -> 67.7 | Improved |
| use LLaVA-UHD adaptive partition instead of fixed 2x2 slices (ablation) | LLaVA-UHD with MLP · TextVQA | accuracy: 66.1 -> 66.9 | Improved |
| add spatial schema (separators for slice positions) (ablation) | LLaVA-UHD · VQAv2 | accuracy: 79.3 -> 81.7 | Improved |
| add spatial schema (separators for slice positions) (ablation) | LLaVA-UHD · TextVQA | accuracy: 65.4 -> 67.7 | Improved |
| update ViT during pretraining as well as instruction tuning (ablation) | LLaVA-UHD · TextVQA | accuracy: 67.7 -> 62.6 | Hurt |
| update ViT during instruction tuning rather than keeping frozen (ablation) | LLaVA-UHD · TextVQA | accuracy: 65.7 -> 67.7 | Improved |
| native-resolution slicing instead of fixed 336x336 input | LLaVA-1.5 · TextVQA | accuracy: 61.3 -> 67.7 | Improved |
| extreme aspect ratio images with LLaVA-UHD vs LLaVA-1.5 | LLaVA-UHD · VQAv2 (extreme aspect ratio) | accuracy: LLaVA-1.5 74.6 (-5.4) vs LLaVA-UHD 81.4 (-0.3) | Improved |
| replace MLP with resampler on ScienceQA (ablation) | LLaVA-UHD · GQA | accuracy: 65.4 (MLP) vs 65.2 | Mixed |
Later papers that compared with this method
1 later accepted paper reports LLaVA-UHD-v2* as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- When Large Vision-Language Model Meets Large Remote Sensing Imagery: Coarse-to-Fine Text-Guided Token Pruning (ICCV 2025 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- Monkey — Monkey: Image Resolution and Text Label Are Important Things for Large Multi-modal Models (CVPR 2024) · better on 0 of 1 shared result
- SPHINX-2k — "SPHINX: A Mixer of Weights, Visual Embeddings and Image Scales for Multi-modal Large Language Models" (ECCV 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- LLaVA-Mini: Efficient Image and Video Large Multimodal Models with One Vision Token (ICLR 2025 · similarity 0.92)
- iLLaVA: An Image is Worth Fewer Than 1/3 Input Tokens in Large Multimodal Models (ICLR 2026 · similarity 0.90)
- LLaVA-Interleave: Tackling Multi-image, Video, and 3D in Large Multimodal Models (ICLR 2025 · similarity 0.90)
- Improved Baselines with Visual Instruction Tuning (CVPR 2024 · similarity 0.89)
- Divide, Conquer and Combine: A Training-Free Framework for High-Resolution Image Perception in Multimodal Large Language Models (AAAI 2025 · similarity 0.88)
- Cambrian-1: A Fully Open, Vision-Centric Exploration of Multimodal LLMs (NeurIPS 2024 · 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.
