AI Applications

AI Image Processing

Image intelligence repair and resolution enhancement

We model the artist's aesthetic mathematics through AI

1. Input image

2. Identification: The artificial intelligence system automatically recognizes unclear image content.  

3. AI "Brain Supplement" image content

 

Detailing with "spliced generation network" and "fixed point generation network"

4. Discriminator Multi-Round Optimization

 

The discriminator identifies the pros and cons of the finished product through human visually sensitive features and performs multiple rounds of optimization.

AI Image Processing Points

Aesthetic Modelling

 

Modeling the aesthetic mathematics of humans and artists, creating objective functions such as ConcatGAN (splicing generation network) and InvarianceGAN (fixed point generation network)

Discriminator optimization

 

n response to the characteristics of the film and television industry, the traditional CNN data was changed to a multiple cycle at a time, and the computing power was exchanged for higher image processing quality. Created a new model structure, LoopNet (circular network), Discriminator assisted final descent (discriminator-assisted convergence) to enable the network to improve output quality through multiple rounds of optimization

Innovate multiple basic neural network components

 

Used to make up for the shortcomings of existing neural network components, especially convolutional networks. This includes local gradient statistics extraction.

Original World Prior Memory

 

It is a way of preserving prior knowledge in neural networks, and it solves the problem of total knowledge limit due to the limitation of network computing being limited by objective computational power.

SD Video → 2K  (processing 26 '45")

Remove the original picture scene effect

Removing deformed moir and sawtooth

Filling of picture details

《血色浪漫》、《亮剑》

Increase resolution: 720*576→1920*1080

Improve picture detail