Ali vision AI training camp-day01

Visual production technology

One . Definition and classification ：

1.1. Definition ： Through one / A series of visual processes , Produce new visual expressions .

 Produce ： Images and videos that people or machines can perceive , Not labels or features ;
 requirement ： new , Different from input 

1.2. classification ：

1.3. Common infrastructure

1.4. Five key dimensions

• Meet the vision / Aesthetic expression
• Conform to semantics / Content logic
• Ensure the richness of the results
• Provide the user with the desired grip 【 controllable 】
• Bring users / Commercial value 【 You can use 】
  

Two . Understand carefully

2.1. Split matting

• distinguish ： Know what it is ？
  

• testing ： distinguish + Know where
  

• Division ： distinguish + testing + Know what each pixel is
  

2.2. Split matting — difficulty

• Complex background
• Fine hair button
• Transparent material
• Occlusion
• The edges are reversed
• Multiscale / The goal is

There is a serious shortage of data , High marking cost

2.3. Split matting – Their thinking

1. Disassembly of complex problems ： crude mask Design + accurate matting
2. Rich data samples ： Design image mask The first mock exam

2.4. Split matting – Model framework

2.5. Split matting – People are like cutouts

2.6. Split matting – Object matting expansion

2.7. Split matting —— Scene matting

3、 ... and . Visual generation

3.1 Visual generation – Framework process

3.2. Visual generation – E-commerce design

• Make a picture according to the picture
  
• Personalized design
  

3.3. Visual generation – Luban scene intelligent art

• IN： Marketing scenarios + raw materials
• OUT： Marketing pictures
• Call mode ：API
  

3.4. Visual generation –AlibabaWood

One click to generate short video ：AlibabaWood

3.5. Visual generation – Framework process

Four . Visual editing – Substitute stealthily

4.1. Video implantation

• At present, most advertisements have nothing to do with content , Not very user friendly , Video Embedding means can embed advertisements into the video , Enhance user experience
  
  

4.2. Visual editing —— Implant position detection and positioning

4.3. Visual editing —— Dynamically detect segmentation

• Separate
• Occlusion relationship, etc
  

4.4. Visual editing —— Video content erasure

4.5. Visual editing —— Text erasure

4.6. Visual editing —— Image size changes

5、 ... and . Visual enhancement —— Repair the old as new

5.1. Visual enhancement —— Video enhancement

5.2. Visual enhancement —— Face restoration enhancement

5.3. Visual enhancement —— Style transfer

5.4. Visual enhancement —— Color development

6、 ... and . Intelligent open platform