Games101 review: rasterization

• MVP What to do after the transformation ？ Naturally, the transformed view is mapped to the screen space .
• In the end Mapping to Screen space Up .
• When all triangles are on the screen , Then draw the triangle , This process is called Rasterize .
• utilize The center of the pixel （ Whether it is in a triangle ）, The function of triangle visibility is sampling .
• Processing anti aliasing （ Anti-Aliasing ）AntiAliasing

Catalog

1. Viewport transform

1.1 Define the screen 、 Screen space

1.2 Viewport transform ： Transform the standard cube coordinates into screen space

2. Rasterize

2.1 triangle ： The most basic element

2.2 sampling ： Discretization of continuous functions , On the center of the pixel (x+0.5,y+0.5) sampling

2.2.1 AABB Bounding box （ Axis alignment bounding box ）： The method of finding the optimal surrounding space of the discrete points

3. Anti aliasing

3.0 Several kinds of graphics Sampling Artifacts

3.1 The reason for the distortion ： The sampling frequency cannot be synchronized with the signal frequency

3.2 frequency domain （Frequency Domain）

3.2.1 The Fourier transform （Fourier Transform）： Time domain --> frequency domain

3.2.2 High frequency signals require high frequency sampling , Some visualization examples  

3.2.3 Frequency domain diagram ： High pass filtering 、 Low pass filtering 、 A band pass filter

3.2.4 Convolution operation ： Addition in time domain = Multiplication in frequency domain

3.2.5 sampling = Repeat the content in the frequency domain  

3.2.6 Anamorphosis = Content mixing in frequency domain , That is, there are overlapping parts on the function

3.2.7  Anti aliasing = Limit the mixing of content in the frequency domain , The high-frequency part is removed by filtering before sampling

3.2.8 From the above, we can get ： The anti aliasing effect can be achieved by convolution in a certain pixel area , Take the mean

3.3 Several anti aliasing techniques

3.3.1 MSAA（ Multisampling anti aliasing ）： Improve resolution , Calculate the average

3.3.2 FXAA(Fast Approximate AA)： Fast approximate anti aliasing  

3.3.3 TAA(Temporal AA)： Time domain anti aliasing

3.3.4 DLSS(Deep Learning Super Sampling)： Deep learning anti aliasing

4. Visibility and occlusion (Visibility/Occlusion)

4.0 Painter algorithm Painter's Algorithm

4.1 Z-Buffering Deep buffer , stay GPU Hardware implementation

---

1. Viewport transform

1.1 Define the screen 、 Screen space

1.2 Viewport transform ： Transform the standard cube coordinates into screen space

2. Rasterize

2.1 triangle ： The most basic element

• Guaranteed to be a plane
• It is easy to define the inside and outside of a triangle （ By cross multiplication ）
•   The attributes of any point in the triangle can be obtained by interpolation

2.2 sampling ： Discretization of continuous functions , On the center of the pixel (x+0.5,y+0.5) sampling

Sampling can be done in time , You can also sample at a location

 

2.2.1 AABB Bounding box （ Axis alignment bounding box ）： The method of finding the optimal surrounding space of the discrete points

Collision detection AABB Bounding box _ Star dot blog -CSDN Blog _aabb Bounding box

3. Anti aliasing

3.0 Several kinds of graphics Sampling Artifacts

3.1 The reason for the distortion ： The sampling frequency cannot be synchronized with the signal frequency

3.2 frequency domain （Frequency Domain）

 

 

3.2.1 The Fourier transform （Fourier Transform）： Time domain --> frequency domain

Use Fourier series to approximate the original function ：

3.2.2 High frequency signals require high frequency sampling , Some visualization examples  

 

3.2.3 Frequency domain diagram ： High pass filtering 、 Low pass filtering 、 A band pass filter

High pass filtering ： Only keep the high frequency part , Visually, the boundary is preserved , Because the boundary is the most dramatic part of the image .

Low pass filtering ： Only the low frequency part is reserved , Visually, it is fuzzy , It can be understood as removing obvious boundary parts . 

 

3.2.4 Convolution operation ： Addition in time domain = Multiplication in frequency domain

thus , There are two ways to realize convolution ：

3.2.5 sampling = Repeat the content in the frequency domain  

3.2.6 Anamorphosis = Content mixing in frequency domain , That is, there are overlapping parts on the function

3.2.7  Anti aliasing = Limit the mixing of content in the frequency domain , The high-frequency part is removed by filtering before sampling

So this is why the principle of aliasing can be solved to a certain extent by filtering ：

3.2.8 From the above, we can get ： The anti aliasing effect can be achieved by convolution in a certain pixel area , Take the mean

 

3.3 Several anti aliasing techniques

3.3.1 MSAA（ Multisampling anti aliasing ）： Improve resolution , Calculate the average

For example, every previous point is divided into 1×1 The pixel is increased to 2×2 Pixels of

Then take samples , And calculate the average

3.3.2 FXAA(Fast Approximate AA)： Fast approximate anti aliasing  

Find the jagged image boundary by image matching , And replace it with a borderless image , Very fast , Independent of sampling , Use image method

3.3.3 TAA(Temporal AA)： Time domain anti aliasing

Reuse the information of the previous frame , amount to MSAA The corresponding samples are distributed in time .

3.3.4 DLSS(Deep Learning Super Sampling)： Deep learning anti aliasing

Some information will be lost during upsampling （ Serrate ）, Insufficient sampling rate , Using deep learning techniques “ guess ” Show the missing details .

4. Visibility and occlusion (Visibility/Occlusion)

4.0 Painter algorithm Painter's Algorithm

 

4.1 Z-Buffering Deep buffer , stay GPU Hardware implementation

• Access the pixel depth value nearest to us in a Buffer in , It's called deep cache ,Z-Buffer
• Cache the information after mixing the pixels closest to us , be called frame buffer

z-buffer Algorithm ：