Gut-Halu: Hallucination Benchmark for Gastrointestinal Image Analysis

Annotated files for a benchmark assessing hallucination in large vision-language models applied to gastrointestinal image analysis. The dataset supports the paper 'Hallucination-Aware Multimodal Benchmark for Gastrointestinal Image Analysis with Large Vision Language Models'. It was created by sandesh-pokhrel and last updated on September 5, 2025.

Use Cases

• Benchmarking model hallucination in medical image analysis based on the described hallucination-aware evaluation framework.
• Training or fine-tuning vision-language models for gastrointestinal diagnostics based on the provided annotations.
• Researching the reliability of AI-generated text descriptions for medical images based on the benchmark's focus.

Strengths

• Dataset is explicitly designed for a published academic benchmark on a specific AI problem (hallucination in VLMs).
• Source images are from a known, publicly available medical imaging dataset (Kvasirv2).
• Last update was recent (2025-09-05).

Limitations

• Row count, file formats, and column-level documentation are unknown.
• Images are not included and must be downloaded separately from an external source.
• License information is not provided.

Provenance

Source

Annotations created by the paper authors; original images from the Kvasirv2 dataset.

Collection Method

Likely manual or semi-automated annotation of images from the Kvasirv2 collection for a specific research benchmark.

Freshness

Last updated 2025-09-05 21:07:26