跳到主要内容
Learn Motion
探索课程产品能力价格

下载 Learn Motion

用手机扫描二维码,下载 App。

App Store

扫码从 App Store 下载 Learn Motion打开商店页面

Google Play

扫码从 Google Play 下载 Learn Motion打开商店页面
登录
Learn Motion

How AI Sees: A Conceptual Guide to Image and Video Understanding

1What a Machine Actually Sees2From Pixels to Patterns: Features3How Deep Networks Learn to See4Beyond Labels: Locating and Describing What Is Seen5Adding Time: Understanding Video6How These Systems Learn and How We Judge Them
Beyond Labels: Locating and Describing What Is Seen

One Label for the Whole Image Is Not Enough

1 / 4
Look at the left side of the diagram: the whole image goes in, and one label comes out. That is classification, and it works fine when there is one obvious subject. Now look at the right side. The same image produces a list — each row names an object and points to where it is. The key difference is not accuracy; it is the shape of the answer. A single label cannot describe a scene with several objects, because there is no place to put the second object. Detection solves that by making the output a list instead of a single value.
0:00 / 0:00

A classifier takes an entire image and produces a single label, such as "dog" or "street scene." The label describes the dominant content of the whole frame. Detection changes the question. Instead of one answer for the whole image, the model must produce a set of answers, each naming an object and stating where that object sits. In a photo with two dogs and a bicycle, classification can only say "dog" or "bicycle" — it cannot report both, and it cannot say that one dog is on the left and the bicycle is in the middle. Detection can. The output of a detector is a list, and each entry in the list contains a class label, a location, and a confidence score. The location is what classification never produced, and producing it is the core new capability of this chapter.

上一页1 / 4下一页

Learn Motion

生成课程,认真学会。

运营方:Wuhan Daoyin Technology Co., Ltd.

联系我们: [email protected]
隐私政策用户服务协议

© 2026 Learn Motion