Research map / Vision-language and multimodal / Human and embodied modeling
BROS: A Pre-trained Language Model Focusing on Text and Layout for Better Key Information Extraction from Documents
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 relative spatial encoding and area-masked pre-training rather than LayoutLM's components (ablation) | LayoutLM†_BASE · FUNSD EE | F1: 76.89 -> 80.58 | Helped |
| Add BROS relative positional encoding (ablation) | LayoutLM†_BASE · FUNSD EE | F1: 76.89 -> 78.84 | Helped |
| Add BROS pre-training objectives (ablation) | LayoutLM†_BASE · FUNSD EE | F1: 76.89 -> 78.44 | Helped |
| Use BROS relative spatial encoding instead of absolute-position encoding (ablation) | LayoutLM†_BASE · FUNSD EL | F1: 47.22 -> 65.24 | Helped |
| Use BROS relative spatial encoding instead of LayoutLMv2's relative encoding (ablation) | LayoutLM†_BASE · FUNSD EL | F1: 53.57 -> 65.24 | Helped |
| Replace the BROS spatial encoding with LayoutLMv2's encoding (ablation) | BROS_BASE · SciTSR EL | F1: 99.28 -> 99.28 | No clear effect |
| Use BROS rather than LayoutLMv2 with the large model | FUNSD EE · FUNSD EE | F1: 83.59 -> 84.52 | Helped |
| Use BROS rather than LayoutLMv2 with the large model | SciTSR EL · SciTSR EL | F1: 99.76 -> 99.58 | Hurt |
| Use BROS rather than LayoutLMv2 on permuted document text | p-FUNSD EE · p-FUNSD | F1: 40.77 -> 76.94 | Helped |
| Use BROS rather than LayoutLMv2 on permuted document text | p-SciTSR EL · p-SciTSR | F1: 99.45 -> 99.33 | Hurt |
| Use BROS rather than LayoutLM on FUNSD with only text and layout | FUNSD EE · FUNSD | F1: 80.54 -> 83.05 | Helped |
| Use BROS rather than LayoutLM on FUNSD with only text and layout | FUNSD EE · FUNSD | F1: 78.95 -> 84.52 | Helped |
Closest papers by meaning
- Relation-Rich Visual Document Generator for Visual Information Extraction (CVPR 2025 · similarity 0.87)
- A Simple yet Effective Layout Token in Large Language Models for Document Understanding (CVPR 2025 · similarity 0.86)
- Read, Parse, Describe: Unified Document Parsing with Visual Element Description Generation (NeurIPS 2026 · similarity 0.85)
- DocLayLLM: An Efficient Multi-modal Extension of Large Language Models for Text-rich Document Understanding (CVPR 2025 · similarity 0.84)
- LDP: Generalizing to Multilingual Visual Information Extraction by Language Decoupled Pretraining (AAAI 2025 · similarity 0.84)
- DocKylin: A Large Multimodal Model for Visual Document Understanding with Efficient Visual Slimming (AAAI 2025 · similarity 0.84)
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.