Research map / Vision-language and multimodal / Multimodal LLMs
Uni-DocRobust: Universal Plug-and-Play Robustness Enhancement for Multi-modal LLMs via Feature Restoration
Yuxuan Zhou, Baole Wei, Xingjian Hu, Haowei Chen, Yu Li, Xingyue Lin, Liangcai Gao, Zhi Tang
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes Uni-DocRobust, a plug-and-play feature-restoration module that improves MLLM robustness to real-world document degradations (noise, blur, low resolution).
Problem
MLLMs degrade substantially on document understanding under real-world degradations such as noise, blur, and low resolution, due to scarce large-scale aligned training data and a lack of transferable restoration solutions across models.
Approach
A universal, plug-and-play robustness module that restores degraded features so it can transfer across MLLMs without retraining the base model; details beyond the abstract are not available.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Uni-DocRobust feature restoration module | InternVL3.5-2B · DocRobust-VQA ChartQA (corrupted) | accuracy: 35.71 -> 59.45 | Improved |
| add Uni-DocRobust | LLaVA1.5-7B · WildDoc | accuracy: 28.71 -> 34.25 | Improved |
| add Uni-DocRobust instead of LoRA fine-tuning | LLaVA1.5-7B · DocRobust-VQA ChartQA | accuracy: 41.95 -> 42.61 | Improved |
| add LoRA fine-tuning on top of Uni-DocRobust | InternVL3.5-2B · DocRobust-VQA ChartQA | accuracy: 59.45 -> 61.55 | Mixed |
| add Uni-DocRobust under white-box adversarial attack | InternVL3.5-2B · Adversarial ChartQA | accuracy: 12.72 -> 17.00 | Improved |
| use feature-level restoration instead of pixel-level DocRes | InternVL3.5-2B · DocRobust-VQA ChartQA | accuracy: 28.76 -> 59.45 | Improved |
| add Phase I multi-teacher distillation (vs random-init, full budget) (ablation) | Uni-DocRobust · DocRobust-VQA average | avg score: 61.06 -> 62.72 | Improved |
| add Phase I distillation (vs random-init under fast adaptation) (ablation) | Uni-DocRobust · DocRobust-VQA average | avg score: 52.29 -> 62.72 | Improved |
| use high-capacity Uni-DocRobust instead of standard low-capacity adapter (ablation) | InternVL3.5-2B · DocRobust-VQA average | avg score: 58.36 -> 62.72 | Improved |
| increase number of teacher encoders in Phase I (one to three) (ablation) | Uni-DocRobust · DocRobust-VQA average | avg score: 55.88 -> 62.72 | Improved |
| use InternViT teacher instead of CLIP teacher (ablation) | Uni-DocRobust · DocRobust-VQA average | avg score | Improved |
| scale base MLLM from 2B to 14B and add Uni-DocRobust | InternVL3.5-14B · DocRobust-VQA ChartQA | accuracy: 50.32 -> 64.79 | Improved |
Closest papers by meaning
- Robust-U1: Can MLLMs Self-Recover Corrupted Visual Content for Robust Understanding? (ICML 2026 · similarity 0.90)
- Res-Bench: Benchmarking the Robustness of Multimodal Large Language Models to Dynamic Resolution Input (AAAI 2026 · similarity 0.88)
- Machine Vision Therapy: Multimodal Large Language Models Can Enhance Visual Robustness via Denoising In-Context Learning (ICML 2024 · similarity 0.88)
- SPHINX-X: Scaling Data and Parameters for a Family of Multi-modal Large Language Models (ICML 2024 · similarity 0.88)
- AlignVLM: Bridging Vision and Language Latent Spaces for Multimodal Document Understanding (NeurIPS 2025 · similarity 0.88)
- Improving Visual Token Reduction via Rectifying Distortions for Efficient Multimodal LLM Inference (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.
