PixelAnnotationTool

Why we included this project

Building a segmentation dataset by hand is slow, and this tool cuts the worst of that work down. You paint rough brush strokes over the regions you want to label, and an OpenCV watershed algorithm expands those markers into full pixel masks, so you get dense per-pixel ground truth without tracing every boundary. When a mask comes out wrong, you draw new markers on the bad spots and rerun the segmentation rather than starting over. It's a Qt desktop app with prebuilt releases for Linux, macOS, and Windows, so annotators never have to write code or build anything themselves. For teams producing training data for semantic segmentation, that makes it a practical way to get dense labels faster than manual outlining alone.