ScreenParse: Large-Scale Dataset for Complete Screen Parsing

ScreenParse is a large-scale dataset for complete screen parsing, providing dense annotations of user interface elements. The dataset was created by the docling-project and updated in May 2026. ScreenParse v2 contains 1,447,100 high-quality training screenshots with varied viewport resolutions and leaf-element annotations.

Use Cases

• Train computer vision models for UI element detection based on dense annotations.
• Develop automated screen readers or accessibility tools based on parsed screen content.
• Benchmark layout understanding algorithms based on varied viewport resolutions.
• Improve annotation quality for UI datasets based on leaf-element annotations that reduce noise.

Strengths

• Contains 1,447,100 high-quality training screenshots.
• Includes varied viewport resolutions, suggesting diversity in screen sizes.
• Version 2 features leaf-element annotations designed to reduce annotation noise.

Limitations

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

Provenance

Source

docling-project

Collection Method

Likely gathered from automated or semi-automated screenshot capture and annotation.

Time Range

null

Freshness

Last updated 2026-05-22 12:59:25.

Geography

null