Research map / Vision-language and multimodal / Multimodal LLMs
TextMonkey: An OCR-Free Large Multimodal Model for Understanding Document
Yuliang Liu, Biao Yang, Qiang Liu, Zhang Li, Zhiyin Ma, Shuo Zhang, Xiang Bai
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 zero initialization to shifted window attention (ablation) | TextMonkey · ChartQA | accuracy: 57.6 -> 58.2 | Improved |
| add shifted window attention (cross-window connections) (ablation) | TextMonkey · DocVQA / TextVQA | accuracy: 62.6 -> 64.1 DocVQA; 62.4 -> 64.8 TextVQA; 45.9 -> 46.0 SROIE | Improved |
| add token resampler on top of window attention (ablation) | TextMonkey · SROIE / DocVQA / TextVQA | accuracy: 46.0 -> 47.0 SROIE; 64.1 -> 64.3 DocVQA; 64.8 -> 65.9 TextVQA | Improved |
| use significant-token filter instead of random queries (ablation) | Token Resampler · SROIE / DocVQA / TextVQA | accuracy: 32.9 -> 47.0 SROIE; 46.7 -> 64.3 DocVQA; 59.5 -> 65.9 TextVQA | Improved |
| add resampler after token filtering instead of dropping tokens (ablation) | Token Resampler · SROIE / DocVQA / TextVQA | accuracy: 44.9 -> 47.0 SROIE; 63.5 -> 64.3 DocVQA; 62.5 -> 65.9 TextVQA | Improved |
| sort selected tokens by order (ablation) | Token Resampler · DocVQA | accuracy: 62.1 -> 64.3 | Improved |
| raise resolution from 896 to 1344 without token compression (ablation) | TextMonkey · DocVQA | accuracy: 64.1 -> 54.9 | Hurt |
| raise resolution to 1344 with 1024 kept tokens vs 896 with 512 (ablation) | TextMonkey · InfoVQA / DocVQA | accuracy: 28.2 -> 31.4 InfoVQA; 64.3 -> 64.5 DocVQA; 65.9 -> 65.1 TextVQA; 47.0 -> 46.0 SROIE | Mixed |
| require model to output answer position (ablation) | TextMonkey · DocVQA / SROIE | accuracy: 64.3 -> 64.5 DocVQA; 47.0 -> 47.2 SROIE | Improved |
| require model to output answer position (ablation) | TextMonkey · ChartQA / InfoVQA | accuracy: 58.2 -> 57.8 ChartQA; 28.2 -> 27.7 InfoVQA | Hurt |
| use point instead of rectangle position representation (ablation) | TextMonkey · DocVQA / SROIE | accuracy: 64.3 -> 65.0 DocVQA; 47.0 -> 47.9 SROIE | Improved |
| use polygon instead of rectangle position representation (ablation) | TextMonkey · DocVQA / SROIE | accuracy: 64.3 -> 64.0 DocVQA; 47.0 -> 47.2 SROIE | Hurt |
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.
- Monkey — Monkey: Image Resolution and Text Label Are Important Things for Large Multi-modal Models (CVPR 2024) · better on 3 of 3 shared results
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
- OCRBench v2: An Improved Benchmark for Evaluating Large Multimodal Models on Visual Text Localization and Reasoning (NeurIPS 2025 · similarity 0.90)
- DocLayLLM: An Efficient Multi-modal Extension of Large Language Models for Text-rich Document Understanding (CVPR 2025 · similarity 0.88)
- DocKylin: A Large Multimodal Model for Visual Document Understanding with Efficient Visual Slimming (AAAI 2025 · similarity 0.88)
- Hierarchical Visual Feature Aggregation for OCR-Free Document Understanding (NeurIPS 2024 · similarity 0.87)
- OmniCorpus: A Unified Multimodal Corpus of 10 Billion-Level Images Interleaved with Text (ICLR 2025 · similarity 0.87)
- Docopilot: Improving Multimodal Models for Document-Level Understanding (CVPR 2025 · 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.
