Skip to content
Free public beta until December 4, 2026: every course is open, no card needed. After that, the first chapter of each course stays free and full access is a one-time $39 ($19 for up to 100 founding members chosen for helpful feedback). Your progress is kept. How the beta works

Every way to segment an image

From a fixed, trained label set to a live point or text prompt -- which segmentation approach actually fits the task in front of you.

Start with what the task actually needs distinguished. Does every object of the same class need its own separate mask, or is one merged region per class enough? Is the label set fixed in advance and worth training for, or does the task need to segment a concept only named at inference time, with no retraining?

Fixed, trained label set

See section →
Labeling every pixel in an image by class, with no notion of separate object instances.

Every pixel gets a class label; same-class objects merge into one region.

See section →
Labeling every pixel by class and separating each individual object instance from every other.

Adds a separate mask per object instance on top of the class label.

Term

Mask R-CNN

See section →
A two-stage detection network extended with a branch that predicts a segmentation mask for each detected object.

The standard two-stage architecture that actually produces instance segmentation's per-object masks.

Open-ended, prompted at inference time

See section →
Segmenting an object from a point, box, or text prompt supplied at inference time, rather than a fixed trained label set.

A point, box, or text prompt segments whatever it indicates, with no fixed label set to retrain against.

See section →
Segmenting every instance of a concept the model was never specifically trained to segment.

Extends a single prompted object to "every instance of this concept" from a description alone.

Graph settings

Calculator