Dry goods | Introduction to target detection, that's enough (more finished)

Target recognition: classify multiple targets in the whole picture and structure the picture information into category information. For example, in animal identification, if there are many cats and birds in a picture, they will be classified into multiple categories of cats and birds.

Target detection: On the basis of target recognition, we should not only know which category each target belongs to, but also know its position in the image, that is, the coordinates of the boundary. In addition, there are usually more targets in graphics than in target recognition.

Target segmentation: target segmentation is to detect the target at pixel level, that is, not to frame the target with a border, but to know which pixel the target belongs to. Target segmentation is divided into semantic segmentation and instance segmentation.

Semantic segmentation means that you only need to know which category the segmented object belongs to. For example, in animal target segmentation, if there are many cats in a picture, it is only necessary to classify them as birds after the target segmentation.

Instance segmentation refers to the need to know which instance the segmented target belongs to. For example, in animal object segmentation, if there are many cats in a picture, they need to be identified as cats 1, cats 2, cats 3 and so on after object segmentation.

Target tracking: In video stream, using time series information, not only the target should be detected, but also the target should be continuously detected in subsequent video frames, and it should also adapt to the scale change of the target.

2. The relationship among target recognition, target detection, target segmentation and target tracking.

① Target recognition is the basis of target detection.

② Target segmentation is pixel-level target detection.

③ Target detection is the basis of target tracking.

3. The difference between target recognition, target detection, target segmentation and target tracking.

① Target recognition focuses on the whole information of the whole picture, while target detection only focuses on the specific target in the picture. For example, pedestrian detection only pays attention to pedestrians and not other objects.

② Target segmentation is pixel-level, but target detection is not.

③ The scene of target tracking application is video streaming, while others are not.