KIE-HVQA: Data for Mitigating OCR Hallucinations in Multimodal LLMs

KIE-HVQA is a dataset supporting research on mitigating Optical Character Recognition hallucinations in multimodal large language models. The dataset was created by bytedance-research and is associated with a paper accepted by the NeurIPS 2025 Main Conference. The data likely contains multimodal document samples for evaluating and improving OCR integration in vision-language models.

Use Cases

• Benchmarking OCR hallucination detection based on multimodal document samples
• Training models for improved document understanding based on integrated textual and visual data
• Evaluating the robustness of multimodal LLMs based on OCR error scenarios

Strengths

• Associated with a peer-reviewed paper accepted by NeurIPS 2025
• Created by a research organization (bytedance-research)
• Focuses on a specific, current research problem in multimodal AI

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download
• Column-level documentation is absent; field semantics must be inferred after download
• Row count and file formats are unknown, which may limit suitability assessment

Provenance

Source

bytedance-research

Collection Method

Created for the research paper 'Seeing is Believing? Mitigating OCR Hallucinations in Multimodal Large Language Models'

Time Range

null

Freshness

Last updated 2025-09-19 02:54:57

Geography

null