Research map / Vision-language and multimodal / Multimodal LLMs
A Token-level Text Image Foundation Model for Document Understanding
Tongkun Guan, Zining Wang, Pei Fu, Zhengtao Guo, Wei Shen, Kai Zhou, Tiezhu Yue, Chen Duan, Hao Sun, Qianyi Jiang, Junfeng Luo, Xiaokang Yang
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 token alignment at LLM level (ablation) | TokenVL · Full-image text recognition TotalText | edit distance: 0.3592 -> 0.3547 | Improved |
| add token alignment at LLM level (ablation) | TokenVL · Full-image text recognition IIT-CDIP | edit distance: 0.2388 -> 0.1921 | Improved |
| add token alignment at LLM level (ablation) | TokenVL · Full-image text recognition DocGenome | edit distance: 0.2837 -> 0.2806 | Improved |
| add token alignment at LLM level (ablation) | TokenVL · InfoVQA | ANLS: 75.3 -> 76.5 | Improved |
| add token alignment at LLM level (ablation) | TokenVL · ChartQA | accuracy: 86.5 -> 86.6 | No clear effect |
| add token alignment at LLM level (ablation) | TokenVL · DocVQA | ANLS: 93.8 -> 94.2 | Improved |
| add token abstractor instead of simple pooling (ablation) | TokenVL · DocVQA | ANLS: 93.1 -> 93.8 | Improved |
| add token abstractor instead of simple pooling (ablation) | TokenVL · ChartQA | accuracy: 86.5 -> 86.5 | No clear effect |
| replace CLIP-L with TokenFD, zero-shot segmentation | text segmentation · zero-shot average over TextSeg, TotalText, HierText | average: 15.55 -> 32.61 | Improved |
| increase TokenFD input resolution from 448px to 1024px | TokenFD zero-shot segmentation · zero-shot average | average: 32.61 -> 34.59 | Improved |
| increase CLIP-L resolution 336px to 1024px | CLIP-L zero-shot segmentation · zero-shot average | average: 15.55 -> 15.81 | No clear effect |
| replace InternViT2.5 with TokenFD as frozen VFM | Vicuna-7B VQA · four VQA benchmarks average | ANLS average: 71.25 -> 72.21 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MiniMonkey — Mini-Monkey: Alleviating the Semantic Sawtooth Effect for Lightweight MLLMs via Complementary Image Pyramid (ICLR 2025) · 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
- TokenFlow: Unified Image Tokenizer for Multimodal Understanding and Generation (CVPR 2025 · similarity 0.89)
- From Pixels to Tokens: Byte-Pair Encoding on Quantized Visual Modalities (ICLR 2025 · similarity 0.89)
- A More Word-like Image Tokenization for MLLMs (CVPR 2026 · similarity 0.88)
- What Do Visual Tokens Really Encode? Uncovering Sparsity and Redundancy in Multimodal Large Language Models (CVPR 2026 · similarity 0.88)
- DocKylin: A Large Multimodal Model for Visual Document Understanding with Efficient Visual Slimming (AAAI 2025 · similarity 0.88)
- Unified Language-Vision Pretraining in LLM with Dynamic Discrete Visual Tokenization (ICLR 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.
