Deep learning is divided into 4 major tasks for image detection:
From the perspective of the task requirements itself, as the amount of information in the processing results of the four tasks increases successively, the complexity of the tools required will successively increase, so the difficulty of completing the corresponding tasks will also increase successively.
(1) Image classification a---> (2) Target detection b---> (3) Semantic segmentation c---> (4) Instance segmentation d (in a way of increasing difficulty)
This picture perfectly explains the four differences.
(1) Image classification: (Image Classification ）
The task of object classification requires binary labels indicating whether objects are present in an image。
The task of image classification requires us to label the objects that appear in a certain image. (For example, input a test picture and output the object category in the picture)
(2) Target detection: (or Object detection)
Detecting an object entails both stating that an object belonging to a specified class is present, and localizing it in the image. The location of an object is typically represented by a bounding box.
Object detection achieves two tasks: one is to determine whether an object belonging to a certain class appears in the picture; the other is to locate the object, and the common characterization of positioning is the bounding box of the object. It can be realized: input the test picture, output the detected object category and location.
(3) Semantic segmentation: (or Semantic scene labeling)
The task of labeling semantic objects in a scene requires that each pixel of an image be labeled as belonging to a category, such as sky, chair, floor, street, etc. In contrast to the detection task, individual instances of objects do not need to be segmented。
Semantic labeling/segmentation: This task needs to label each pixel in the image as a certain object category. Different instances of the same object do not need to be segmented separately. For the picture below, label it as human, sheep, dog, and grass. Instead of sheep 1, sheep 2, sheep 3, sheep 4, and sheep 5.
(4) Instance segmentation (Instance segment ）
Instance segmentation is a combination of object detection and semantic segmentation . Relative to the bounding box of object detection, instance segmentation can be accurate to the edge of the object; relative semantic segmentation, instance segmentation can mark different individuals of the same object on the map (sheep 1, sheep 2, sheep 3...)
-----------end-------