Research map / Vision-language and multimodal / Multimodal LLMs
BaFCo: A Document Understanding Benchmark for Complex Bangla Form Comprehension
Abu Tyeb Azad, Ishita Apan, Fahim Ahmed, Sumaiya Karim Katha, Ezharuddin Jubaer, Armun Alam, Pranjal Nandi, Amin Ali, Aman Chadha, Md Mofijul Islam, A K M Mahbubur Rahman
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 coarse 5-class entity set instead of granular 26-class (ablation) | Gemini 3 Pro · BaFCo DLA, high reasoning, zero-shot, IoU@0.3 | mAP: 0.1177 -> 0.2646 | Improved |
| increase reasoning effort from low to high (ablation) | Kimi K2.5 · BaFCo DLA coarse, zero-shot, IoU@0.3 | mAP: 0.0469 -> 0.1336 | Improved |
| increase reasoning effort from low to high (ablation) | Qwen 3.6-Plus · BaFCo DLA coarse, zero-shot, IoU@0.3 | mAP: 0.0330 -> 0.0589 | Improved |
| increase reasoning effort from low to high (ablation) | Gemini 3 Pro · BaFCo DLA granular, zero-shot, IoU@0.3 | mAP: 0.0900 -> 0.1177 | Improved |
| increase reasoning effort from low to high (ablation) | GPT-5.2 · BaFCo DLA coarse, zero-shot, IoU@0.3 | mAP: 0.1800 -> 0.1434 | Hurt |
| increase reasoning effort from low to high (ablation) | Gemini 3 Pro · BaFCo DLA granular, zero-shot, IoU@0.3 | avg IoU: 0.2430 -> 0.2243 | Hurt |
| add chain-of-thought prompting over zero-shot (ablation) | Gemini 3 Pro · BaFCo DLA granular, high reasoning, IoU@0.3 | mAP: 0.1177 -> 0.1134 | No clear effect |
| add chain-of-thought prompting over zero-shot (ablation) | GPT-5.2 · BaFCo DLA granular, high reasoning, IoU@0.3 | mAP: 0.0511 -> 0.0615 | Improved |
| use English forms instead of Bangla forms (ablation) | Gemini 3 Pro · BaFCo KIE | Macro-F1: 0.848 -> 0.828 | Hurt |
| use English forms instead of Bangla forms (ablation) | Claude Opus 4.6 · BaFCo KIE | Macro-F1: 0.678 -> 0.835 | Improved |
| use English forms instead of Bangla forms (ablation) | Kimi K2.5 · BaFCo KIE | Macro-F1: 0.681 -> 0.828 | Improved |
| use English forms instead of Bangla forms (ablation) | GPT-5.2 · BaFCo DLA granular entity set, IoU@0.3 | mAP: 0.09 -> 0.11 | No clear effect |
Closest papers by meaning
- SRFUND: A Multi-Granularity Hierarchical Structure Reconstruction Benchmark in Form Understanding (NeurIPS 2024 · similarity 0.87)
- Docopilot: Improving Multimodal Models for Document-Level Understanding (CVPR 2025 · similarity 0.86)
- MMLONGBENCH-DOC: Benchmarking Long-context Document Understanding with Visualizations (NeurIPS 2024 · similarity 0.86)
- SO-Bench: A Structural Output Evaluation of Multimodal LLM (CVPR 2026 · similarity 0.86)
- MosaicDoc: A Large-Scale Bilingual Benchmark for Visually Rich Document Understanding (AAAI 2026 · similarity 0.85)
- DocLayLLM: An Efficient Multi-modal Extension of Large Language Models for Text-rich Document Understanding (CVPR 2025 · similarity 0.85)
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.
