Research map / Vision-language and multimodal / Human and embodied modeling
WordScape: a Pipeline to extract multilingual, visually rich Documents with Layout Annotations from Web Crawl Data
Maurice Weber, Carlo Siebenschuh, Rory Marshall Butler, Anton Alexandrov, Valdemar Ragnar Thanner, Georgios Tsolakis, Haris Jabbar, Ian Foster, Bo Li, Rick Stevens, Ce Zhang
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 |
|---|---|---|---|
| Pretrain on 10k WordScape samples before finetuning with 25 FUNSD samples | Faster R-CNN · FUNSD | F1 @IoU 0.5: 0.621 -> 0.840 | Improved |
| Pretrain on 50k WordScape samples before finetuning with 50 FUNSD samples | Faster R-CNN · FUNSD | F1 @IoU 0.5: 0.690 -> 0.870 | Improved |
| Pretrain on 50k WordScape samples before finetuning with 100 FUNSD samples | Faster R-CNN · FUNSD | F1 @IoU 0.5: 0.723 -> 0.857 | Improved |
| Pretrain on 100k WordScape samples before finetuning with 149 FUNSD samples | Faster R-CNN · FUNSD | F1 @IoU 0.5: 0.772 -> 0.882 | Improved |
| Pretrain on 5k WordScape samples before finetuning with 75 cTDaR samples | YOLOv8m · ICDAR 2019 cTDaR modern | mAP @ IoU [0.50:0.95]: 0.869 ± 0.008 -> 0.924 ± 0.007 | Improved |
| Pretrain on 10k WordScape samples before finetuning with 150 cTDaR samples | YOLOv8m · ICDAR 2019 cTDaR modern | mAP @ IoU [0.50:0.95]: 0.888 ± 0.011 -> 0.931 ± 0.010 | Improved |
| Pretrain on 10k WordScape samples before finetuning with 300 cTDaR samples | YOLOv8m · ICDAR 2019 cTDaR modern | mAP @ IoU [0.50:0.95]: 0.949 ± 0.006 -> 0.961 ± 0.005 | Improved |
| Pretrain on 10k WordScape samples before finetuning with 600 cTDaR samples | YOLOv8m · ICDAR 2019 cTDaR modern | mAP @ IoU [0.50:0.95]: 0.974 ± 0.003 -> 0.975 ± 0.003 | No clear effect |
| Pretrain on 200k WordScape images rather than random initialization | YOLOv5 · DocLayNet, N_f=1k | mAP @ IoU [0.50:0.95]: 0.299 -> 0.508 | Improved |
| Pretrain on 200k WordScape images rather than PubLayNet | YOLOv5 · DocLayNet, N_f=1k | mAP @ IoU [0.50:0.95]: 0.467 -> 0.508 | Improved |
| Pretrain on 200k WordScape images rather than PubLayNet | YOLOv5 · DocLayNet, N_f=5k | mAP @ IoU [0.50:0.95]: 0.659 -> 0.679 | Improved |
| Pretrain on 200k WordScape images rather than PubLayNet | YOLOv5 · DocLayNet, N_f=20k | mAP @ IoU [0.50:0.95]: 0.720 -> 0.734 | Improved |
Closest papers by meaning
- Docopilot: Improving Multimodal Models for Document-Level Understanding (CVPR 2025 · similarity 0.88)
- DocLayLLM: An Efficient Multi-modal Extension of Large Language Models for Text-rich Document Understanding (CVPR 2025 · similarity 0.87)
- Read, Parse, Describe: Unified Document Parsing with Visual Element Description Generation (NeurIPS 2026 · similarity 0.87)
- MosaicDoc: A Large-Scale Bilingual Benchmark for Visually Rich Document Understanding (AAAI 2026 · similarity 0.87)
- Adaptive Markup Language Generation for Contextually-Grounded Visual Document Understanding (CVPR 2025 · similarity 0.86)
- MPDocBench-Parse: Benchmarking Practical Multi-page Document Parsing (NeurIPS 2026 · similarity 0.86)
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.
