EndoVL: Expert Annotations for Public Gastrointestinal Endoscopic Images

EndoVL provides an expert-annotated annotation layer for 2,709 public gastrointestinal endoscopic images sourced from 9 public datasets. The release includes final annotation workbooks, deidentified expert-level spreadsheets, consensus-support audit tables, metadata, validation outputs, and structured JSON records. The dataset was authored by Peiyao Fu and last updated on May 20, 2026.

Use Cases

• Train medical image segmentation models based on expert-annotated lesion boundaries.
• Develop uncertainty-aware AI models based on the provided uncertainty labels.
• Benchmark annotation quality and inter-rater reliability using the consensus-support audit tables.
• Validate model performance on a curated set of public endoscopic images.

Strengths

• Contains expert-level annotations for 2,709 images, which suggests high-quality labels.
• Includes uncertainty labels, a feature valuable for training robust medical AI.
• Aggregates images from 9 public source datasets, providing a consolidated benchmark resource.
• Provides structured JSON records and validation outputs for reproducibility.

Limitations

• The original source images are not included and must be obtained separately under their respective licenses.
• Row count and specific column-level documentation are unknown, requiring inspection after download.
• The dataset's 4.5 MB size indicates it contains only annotations and metadata, not the image files themselves.

Provenance

Source

Aggregated from 9 public source datasets listed in metadata.

Collection Method

Expert annotation and consensus-support auditing.

Freshness

Last updated 2026-05-20 06:17:14.