Semantic Human Matting accurately estimates foreground objects in images and videos, making it important for editing applications

Matting Human Dataset

The ability to recognize and analyze images and video
Object extraction in images for the purpose of overlaying them on a new background
Highlighting objects in photos to train systems to recognize and interpret them
Using machine learning in graphic design
Data Labeling
Computer Vision
Editing applications
3 500
4 weeks
Project Duration
Matting Example
This dataset is an example of a custom-made project.
Data: selfies of men and women from Europe and Latin America, collected through a crowdsourcing platform
Labeling: high-quality semantic segmentation in two classes (foreground and background)
Format: PNG files, csv file with metadata (information about the country of residence, age and gender)
Optional Services
Data collection: metadata is prepared for each participant and additional images are collected from internal resources or crowdsourcing.
Foreground color correction: adjustment modifies the color and tone of the extracted object to provide a more natural appearance when superimposed on a different background.
Pixel transparency assessment: creating an alpha mask to determine the degree of transparency of an object
Quality selection: for high quality, use detailed pixel-by-pixel annotation in Photoshop. For medium quality, use broken lines (polygon vertices).
Tell us about your project!
See other datasets