Video Compression - System Design

Añadido: 2026-05-28

[00:00:00]If you don't know this, you suck at system design. A 4K video is going to take you four times as many pixels as 1080p. If you're watching it at just 30fps, that means you need a whopping 6 Gbits per second bit rate to support that stream. Now, the average internet in America is like 200 megabits per second. So, how could you support a stream that big? As you likely know, the answer is compression. And we do it in two different ways. The first way is spatial compression, which is per frame.

[00:00:28]And the other one is temporal compression, which goes across frames.

[00:00:32]State-of-the-art video compression will likely use stuff like a forier transform. And that gets pretty complicated. But a simple idea is the idea of say run length encoding. If you have five A's in a row and then six B's after that, why would you write them all out when you could just write 5 A and then 6B? You save half the data and say exactly the same thing. Now, the other one is temporal compression, which is across frames. The main idea is just only say what changes. So if you have a background, pretty much every pixel is the same thing. But you typically have one object throughout the frame, probably you, that's moving around.

[00:01:09]Therefore, you just actually write down what pixels change. And if nothing even changed, then don't bother storing anything. And as you probably know, there's two main types of compression.

[00:01:20]Lossy will actually remove unnecessary data and lossless will just restructure the file and keep it at perfect quality.

[00:01:27]I teach this stuff every day and growing really fast. So follow along to learn