Research map / Vision-language and multimodal / Visual question answering
A Question-Answering Approach to Key Value Pair Extraction from Form-Like Document Images
Kai Hu, Zhuoyuan Wu, Zhuoyao Zhong, Weihong Lin, Lei Sun, Qiang Huo
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 KVPFormer instead of Baseline #2 | LayoutLM-based key-value extraction · FUNSD | F1: 0.7884 -> 0.8223 | Improved |
| Use KVPFormer instead of Baseline #2 | LayoutXLM-based key-value extraction · XFUND multitask, Avg. | F1: 0.9257 -> 0.9442 | Improved |
| Use KVPFormer instead of Baseline #2 | LayoutXLM-based key-value extraction · XFUND zero-shot transfer, Avg. | F1: 0.8954 -> 0.9041 | Improved |
| Add gold entity label embeddings | KVPFormer · FUNSD | F1: 0.8223 -> 0.9086 | Improved |
| Add the Transformer encoder (ablation) | Baseline key-value extraction · FUNSD | F1: 0.7498 -> 0.7632 | Improved |
| Add the Transformer decoder (ablation) | Baseline key-value extraction · FUNSD | F1: 0.7498 -> 0.7767 | Improved |
| Use the encoder-decoder architecture rather than encoder-only or decoder-only (ablation) | KVPFormer components without SCAB or C2F · FUNSD | F1: 0.7632 / 0.7767 -> 0.7825 | Improved |
| Add SCAB (ablation) | Encoder-only KVPFormer · FUNSD | F1: 0.7632 -> 0.7817 | Improved |
| Add SCAB (ablation) | Decoder-only KVPFormer · FUNSD | F1: 0.7767 -> 0.7906 | Improved |
| Add SCAB (ablation) | Encoder-decoder KVPFormer · FUNSD | F1: 0.7825 -> 0.8017 | Improved |
| Add coarse-to-fine answer prediction (ablation) | Encoder-decoder KVPFormer with SCAB · FUNSD | F1: 0.8017 -> 0.8223 | Improved |
Closest papers by meaning
- Rethinking Data Augmentation for Robust Visual Question Answering (ECCV 2022 · similarity 0.85)
- Leveraging Visual Question Answering for Image-Caption Ranking (ECCV 2016 · similarity 0.84)
- An Analysis of Visual Question Answering Algorithms (ICCV 2017 · similarity 0.84)
- Information Maximizing Visual Question Generation (CVPR 2019 · similarity 0.84)
- Separation of Powers: On Segregating Knowledge from Observation in LLM-enabled Knowledge-based Visual Question Answering (CVPR 2025 · similarity 0.84)
- VTQA: Visual Text Question Answering via Entity Alignment and Cross-Media Reasoning (CVPR 2024 · similarity 0.83)
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.
