Overview of neural network method architectures

HDRCNN

UNet-VGG-16 with skip connections neural network.

HDRTVNet

3 sequentially trained networks:

• Custom architecture
• ResNet
• UNet based GAN

ExpandNet

• Neural network with three parallel branches. They differ in the size of the convolution core
• Concatenation followed by convolution processing

DeepHDR

UNet-VGG-like with skip connections neural network.

SingleHDR’20

Network which is divided into two parts: linearization and restoration.

• Linearization: custom architecture
• Restoration: ResNet

FHDR

FHDR architecture:

IntrinsicHDR

The in-camera pipeline modeling LDR image formation and IntrinsicHDR architecture:

ITM25

ITM25 pipeline:

ITMLUT

ITMLUT architecture: