Understanding Classification, Classification with Localization and Object Detection

Image Classification:

In image classification, a supervised machine learning algorithm looks at input image and predicts a class label of that image (binary or multi-class). For example, a binary dog classifier algorithm will predict dog or no-dog class label for a given input image.

Image Classification with Localization:

In image classification with localization, a supervised algorithm is expected to predict class as well as bounding box around object in the image. The term ‘localization’ refers to where the object is in the image. Let's say we are talking about classification of vehicle with localization. In this case, the algorithm will predict class of vehicle along with coordinates of bounding box around vehicle object in the image.

Object Detection:

In object detection problem, we have multiple objects in the image and algorithm is expected to detect all of them and put bounding box around each of them. For instance, if we are doing object detection for autonomous vehicle application we need to detect and localize cars, trucks, motorbikes and pedestrians.

Classification and classification with localization usually have one object in the image. However, in object detection problem, the input image may have multiple objects of different types. These three problem are closely related to each other as we move from classification to classification with localization and to object detection.

In computer vision, object detection has made huge progress over last decade, especially in applications such as face recognition, video surveillance, vehicle detection, online image classification, object detection in manufacturing industry etc.

Got A Data Science Question?

Ask our experts anything about machine learning, analytics or statistics.