Research map / Vision-language and multimodal / Visual question answering
SEA-Vision: A Multilingual Benchmark for Comprehensive Document and Scene Text Understanding in Southeast Asia
Pengfei Yue, Xingran Zhao, Juntao Chen, Peng Hou, Wang Longchao, Jianghang Lin, Shengchuan Zhang, Anxiang Zeng, Liujuan Cao
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 |
|---|---|---|---|
| evaluate on low-resource scripts (Khmer, Lao, Burmese) instead of English | MLLMs on TEC-VQA · SEA-Vision TEC-VQA | accuracy (%): 62.85 -> 8.17 (EN vs MY avg) | Hurt |
| use general MLLM (Gemini2.5-Pro) instead of best expert model dots.ocr | document parsing · SEA-Vision Document Parsing | NED: 0.186 -> 0.159 | Improved |
| use specialized pipeline PaddleOCR-VL instead of general MLLMs on high-resource languages | PaddleOCR-VL · SEA-Vision Document Parsing (EN) | NED: 0.154 -> 0.108 | Improved |
| use Gemini2.5-Pro on low-resource scripts vs PaddleOCR-VL | document parsing · SEA-Vision Document Parsing (KM) | NED: 0.634 -> 0.278 | Improved |
| use expert models (dots.ocr) vs general models on fixed-layout pages | document parsing · SEA-Vision page types (Academic Paper) | NED: dots.ocr 0.141 best on Academic Paper; Gemini 0.123 on Magazine | Mixed |
| use closed-source vs open-source MLLMs | TEC-VQA · SEA-Vision TEC-VQA | accuracy (%): closed avg 37.49 vs open avg 31.34; best Qwen3-VL-32B 40.14 (open) > Gemini 39.49 | Mixed |
| scale to TEC-VQA instead of document parsing | MLLMs · SEA-Vision | accuracy / NED: 32.36 avg TEC-VQA accuracy | Hurt |
Closest papers by meaning
- IndicVisionBench: Benchmarking Cultural and Multilingual Understanding in VLMs (ICLR 2026 · similarity 0.88)
- Cross-Lingual Text-Rich Visual Comprehension: An Information Theory Perspective (AAAI 2025 · similarity 0.88)
- CVQA: Culturally-diverse Multilingual Visual Question Answering Benchmark (NeurIPS 2024 · similarity 0.88)
- MosaicDoc: A Large-Scale Bilingual Benchmark for Visually Rich Document Understanding (AAAI 2026 · similarity 0.88)
- Marten: Visual Question Answering with Mask Generation for Multi-modal Document Understanding (CVPR 2025 · similarity 0.87)
- MDPBench: A Benchmark for Multilingual Document Parsing in Real-World Scenarios (NeurIPS 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.
