Label photos, export COCO and YOLO

Who this annotation tool is for

Training an instance segmentation model needs images in which every object has an exact outline and a class. CutLift works as a COCO annotation tool and a YOLO annotation tool for that job: researchers, students and small teams can label images for machine learning in the browser, without setting up a labelling server or sending images to one. It labels one photo at a time, which suits careful work on hard images.

Fast, accurate masks

Pixel-accurate mask annotation is usually the slow part. AI select turns a click into a mask, Segment everything makes a layer for every object, and Select by name selects every “car” or “person” you type, each on its own layer; all three are Pro. The free tools cover the rest: Quick Selection, Magic Wand, lassos, brushes and the Pen for polygon annotation of hard-edged objects.

The Select and Layer menus speed up cleanup with Fill holes, Remove specks, Keep largest piece, Split into separate pieces, Merge selected layers, and Select tiny pieces to find leftovers under 0.1 % of the image.

Classes and labels

Add class… creates a category, and Assign class to selected layers labels many objects at once. Select all layers of this class gathers them again for checking, and Toggle colour layers by class colours the overlay by category, so a mislabelled object stands out.

What the export contains

File ▸ Export collects everything in one ZIP. The COCO JSON export holds categories, bounding boxes, areas and segmentation as polygons or RLE; YOLO segmentation labels come as one normalised polygon per line with a classes.txt. Masks (hard or soft greyscale), a label map with a colour version and legend, SVG outlines and a Simple JSON file can be added, for the active, selected, visible or all layers.

Masks and Simple JSON are free unless the photo used a Pro tool; COCO, YOLO, label maps and SVG are CutLift Pro Studio exports, $7.99 a month or $47.99 a year, with one free premium download per account. Save a project file to come back and correct labels later.