10.4.1 Segmentation Roadmap: Pixel-Level Regions

Segmentation is finer than detection. Instead of a box, it outputs a mask that says which pixels belong to a class or instance.

See the Mask Workflow First

The main object in this chapter is the mask. The main failure is often boundary quality, tiny objects, occlusion, or class confusion.

Run a Mask IoU Check

This script compares two tiny binary masks.

truth = [
    [1, 1, 0],
    [1, 0, 0],
    [0, 0, 0],
]

pred = [
    [1, 0, 0],
    [1, 1, 0],
    [0, 0, 0],
]

intersection = 0
union = 0
for y in range(3):
    for x in range(3):
        intersection += truth[y][x] == 1 and pred[y][x] == 1
        union += truth[y][x] == 1 or pred[y][x] == 1

print("mask_iou:", round(intersection / union, 3))

Expected output:

mask_iou: 0.5

Segmentation reports should show masks, metrics, and boundary errors, not only a colored overlay.

Learn in This Order

Step	Read	Practice Output
1	Semantic segmentation	Predict one class for every pixel
2	Instance segmentation	Separate different objects of the same class
3	Segmentation practice	Compare masks, IoU/Dice, boundary errors, and failed samples

Evidence to Keep

Keep this page’s proof of learning as a small evidence card:

Input Image

original image and target mask or class map

Prediction

predicted mask, overlay visualization, and boundary examples

Metric

IoU, Dice, per-class score, and boundary failure notes

Failure Check

annotation quality, thin boundary, small region, or class confusion

Expected Output

mask overlay plus segmentation metric summary

Pass Check

You pass this chapter when you can create or inspect a mask, compute a simple overlap metric, and explain one boundary or class-confusion failure.

Check reasoning and explanation

1. A passing answer maps the task to the right visual output: class label, bounding box, mask, OCR text, embedding, or video event.
2. The evidence should include a rendered visual artifact and one metric or qualitative error note.
3. A good self-check names one visual failure mode such as class confusion, missed objects, bad masks, lighting shift, domain shift, or weak annotation quality.