Computer Vision Tasks

Four things you can ask a machine to do with a picture. They sound alike, and the board sets them against each other as options nearly every year — but they are told apart by one simple question.

1The question that decides it

Ask: how much detail comes back?

Just a name → Classification.
A name and a place, for one object → Classification + Localisation.
Many objects found and boxed → Object Detection.
Down to every pixel → Instance Segmentation.

Four tasks, one picture

The same photograph, asked four different questions. What comes back is what tells the tasks apart.

Classification

Assigning an input image one label from a fixed set of categories. It says what is in the picture, and nothing about where.

What comes out: One label for the whole image.

The giveaway in an exam question is how much detail is asked for. Just a name → classification. A name and a place, one object → classification + localisation. Many objects found → object detection. Down to the pixel → instance segmentation.

2The four tasks

Classification

The task of assigning an input image one label from a fixed set of categories. It is one of the core problems in computer vision and, despite its simplicity, has a large variety of practical applications.

Classification + Localisation

Involves both processes: identifying what object is present in the image and, at the same time, where in the image that object is. It is used only for single objects.

Object Detection

The process of finding instances of real-world objects — faces, bicycles, buildings — in images or videos. These algorithms typically use extracted features and learning algorithms to recognise instances of an object category. Commonly used in image retrieval and automated vehicle parking systems.

Instance Segmentation

The process of detecting instances of the objects, giving them a category, and then giving each pixel a label based on that. A segmentation algorithm takes an image as input and outputs a collection of regions, or segments.

3All four in one table

Task	What comes back	How many objects
Classification	One label for the whole image	It does not say
Classification + Localisation	A label and its location	One only
Object Detection	A label and a box for each object found	Many
Instance Segmentation	A label per object, and a category for every pixel	Many

The two that are confused most

Classification + Localisation is for a single object. If the question mentions more than one object, or says “instances”, it is Object Detection. The word single in a question is the examiner pointing at localisation.

4From the board papers

Anita and Surjit are creating an AI application that will classify different types of fruits. The Computer Vision task that will identify the type of fruit and assign a label to it is called ________.
(A) Segmentation (B) Classification (C) Classification + Localization (D) Object DetectionCBSE board exam 2026

1 mark

▶Show the answer

(B) Classification. The question asks only for thetype and a label — nothing about where the fruit is in the picture. No location asked for means no localisation, no detection and no segmentation.

In Computer Vision, which of the following tasks is used for single objects?
(A) Object Detection (B) Classification + Localization (C) Instance Segmentation (D) Non-LocalizationCBSE board exam 2025

1 mark

▶Show the answer

(B) Classification + Localization. “Used only for single objects” is part of its definition. Note that (D) “Non-Localization” is not a real task at all — the paper does invent an option now and then.

______ is the process of finding instances of real-world objects in images or videos.
(a) Instance segmentation (b) Object detection (c) Classification (d) Image segmentationCBSE sample papers 2023–24 and 2024–25

1 mark

▶Show the answer

(b) Object detection. The phrase finding instances of real-world objects is lifted straight from the definition, so learn that wording.

5Check yourself

Quick Check

What is the core task of image classification?

Quick Check

Which task labels every pixel of an object?

Quick Check

A self-driving car finds every pedestrian, cyclist and car in view and boxes each one. This is:

← Previous

👁️ What is Computer Vision?

Pixels, Resolution and Pixel Value 🟦