Build Training Data for Referring Image Segmentation
Mark an object in an image, generate its segmentation mask, and extract a clean cutout — all in your browser. Create high-quality datasets for RIS model training.
How It Works
Three simple steps from raw image to training-ready data.
Mark
Draw a circle, scribble, arrow, or check mark around the target object in the image to indicate what you want to segment.
Segment
The ML model interprets your mark and produces a precise segmentation mask of the intended object.
Extract
Get a clean binary mask and cutout of the segmented object, ready for OCR, LLM processing, or downstream tasks.
Features
Everything you need to build and manage RIS datasets.
Upload images and draw marks with pen tools, stroke styles, colors, and undo support.
All data lives in your browser's IndexedDB. No server needed — your images never leave your device.
Export schema-validated datasets as ZIP files, ready for training ML models.
Combine ZIP datasets from multiple contributors into a single unified training set.
Supported Mark Types
Various stroke types to indicate the target object.
Ready to Build Your Dataset?
Start capturing images and creating annotations right away — everything runs locally in your browser.