Research map / Vision-language and multimodal / Multimodal LLMs
Real5-OmniDocBench: A Full-Scale Physical Reconstruction Benchmark for Robust Document Parsing in the Wild
Cheng Cui, Changda Zhou, Tingquan Gao, Xueqing Wang, ZIYUE GAO, Jing Tang, Yi Liu
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 |
|---|---|---|---|
| use compact specialized PaddleOCR-VL-1.6 (0.9B) instead of Qwen3-VL-235B | document parsing · Real5-OmniDocBench | Overall: 88.90 -> 93.19 | Improved |
| use PaddleOCR-VL-1.6 instead of Gemini-3 Pro | document parsing · Real5-OmniDocBench | Overall: 89.24 -> 93.19 | Improved |
| pipeline tool PP-StructureV3 under Skew instead of Scanning | PP-StructureV3 · Real5-OmniDocBench | Overall: 84.68 -> 37.98 | Hurt |
| pipeline tool Marker-1.8.2 under Skew instead of Scanning | Marker-1.8.2 · Real5-OmniDocBench | Overall: 70.27 -> 41.27 | Hurt |
| Warping instead of Scanning | Deepseek-OCR 2 · Real5-OmniDocBench | Overall: 89.59 -> 66.53 | Hurt |
| Skew instead of Scanning | Qwen3-VL-235B · Real5-OmniDocBench | Overall: 89.43 -> 86.56 | Hurt |
| Skew instead of Scanning | PaddleOCR-VL-1.6 · Real5-OmniDocBench | Overall: 94.74 -> 92.66 | Hurt |
| Skew instead of Scanning | MinerU2.5-Pro · Real5-OmniDocBench | Overall: 92.11 -> 81.26 | Hurt |
| scale up to 235B parameters (Qwen3-VL-235B vs 72B) | Qwen-VL · Real5-OmniDocBench | Overall: 86.92 -> 88.90 | Improved |
| PaddleOCR-VL-1.6 over 1.5 release | PaddleOCR-VL · Real5-OmniDocBench | Overall: 92.05 -> 93.19 | Improved |
Closest papers by meaning
- Towards Real-World Document Parsing via Realistic Scene Synthesis and Document-Aware Training (CVPR 2026 · similarity 0.91)
- DocPTBench: Benchmarking End-to-End Photographed Document Parsing and Translation (NeurIPS 2026 · similarity 0.89)
- OmniDocBench: Benchmarking Diverse PDF Document Parsing with Comprehensive Annotations (CVPR 2025 · similarity 0.89)
- MPDocBench-Parse: Benchmarking Practical Multi-page Document Parsing (NeurIPS 2026 · similarity 0.87)
- MDPBench: A Benchmark for Multilingual Document Parsing in Real-World Scenarios (NeurIPS 2026 · similarity 0.87)
- PP-OCRv5: A Specialized 5M-Parameter Model Rivaling Billion-Parameter Vision-Language Models on OCR Tasks (CVPR 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.
