Screen Content Image and Video Dataset with Pixel-Level Segmentation

A multi-task dataset for screen content image and video coding, containing images and videos composed of synthetic and natural regions. It includes pixel-level segmentation maps distinguishing synthetic from natural content. The dataset was created by Rashid Zamanshoar Heris and is hosted by Harvard Dataverse.

Use Cases

• Train multi-task models for screen content compression using pixel-level synthetic–natural segmentation maps.
• Develop video coding algorithms on screen content videos containing mixed synthetic and natural regions.
• Benchmark segmentation model performance for distinguishing synthetic from natural content in screen images.

Strengths

• Dataset is specifically designed for multi-task learning in screen content coding, a specialized domain.
• Includes pixel-level ground truth segmentation maps for synthetic and natural regions.
• Hosted on the Harvard Dataverse platform, indicating a level of institutional curation.

Limitations

• The dataset size, number of files, and specific video/image resolutions are unknown.
• The composition and balance between synthetic and natural content regions are not detailed.
• No information is provided on the diversity of screen content types or application sources.

Provenance

Source

Harvard Dataverse

Collection Method

Composed of screen content images and videos with generated pixel-level segmentation maps.

Freshness

Last updated on 2026-01-21.