How to works better

How to choose model combination

There is no universal combination, need to be based on the scene and mosaic type, size, flexible selection of different model combinations, and with the choice of denoise, downscale and other functions.
The type of mosaic is very important. For the type of intense flicker and special noise currently there is no effective way to deal with.

1. Basic idea.
   

	work
↓	Super-Resolution
↓	denoise suppress flicker and noise
↓	secondary enhancement (to improve image resolution)
↓	choose whether to denoise again.
↓	call tvai/veai(to improve detail and eliminate noise)
↓	and improve sharpness by UE model.

1. 4X model stage.
   Basic process: TG-DE-(SE)-(DE)-(VEAI)-(UE).
   The disadvantage is that it has too many steps, takes too long, and easily causes image distortions.
   

2. 8X model stage:
   The SuperResolution effect of the 8X model is very obvious and can use Simplified process.
   Basic process: 8X(DA)-(DE)(DS)-(TVAI/VEAI)-(UE)
   For some flicker and abnormal noise types, the basic process is: AF-DS-8X-DE(DS)-(TVAI/VEAI) -(UE)
   

3. Others references:
   DA-DS-8X-(DE)-(TVAI/VEAI)-(UE)
   8X-DS-8X
   

Short	Model/Func
TG	TG-STD,TG-AF1,TG-AF2,TG-8X
DA	TecoGAN-da
8X	Bvpp-8X
AF	TG-AF1,Bvpp-AF
DE	denoise
DS	downscale
UE	UE-4X
*()	*Optional

Does it make sense to reuse the base model?

In the 4X stage, it does not make sense to reuse the same model and does not enhance the actual effect.
In the 8X stage, the reuse of 8X (DA) models can improve the image effect to some extent.