# ZeroDCE

> * Reference \:
> Zero-Reference Deep Curve Estimation for Low-Light Image Enhancement https://github.com/Li-Chongyi/Zero-DCE

## Introduction


ZeroDCE generate a set of curve adjustment parameter to adjust lighting condition of an image. This method can avoid creating artifact or generate overexposure or underexposure region.

## Curve Design

* The curve can be iteratively applied to approximate higher-order curves for more robust and accurate dynamic range adjustment.
* The formula for the curve is designed to be differentiable and monotonous.
* The output image will be reiterated for model training
* Iteratively enhance formula: 𝐿𝐸_𝑛 (𝑥)=𝐿𝐸_(𝑛−1) (𝑥)+𝒜_𝑛 𝐿𝐸_(𝑛−1) (𝑥)(1−𝐿𝐸_(𝑛−1) (𝑥))
* 𝒜←\[−1, 1\], 𝐼(𝑥)←\[0, 1\],𝑛=8
* To avoid overflow truncation, the input pixel value of the enhanced image should be in the normalized rage of \[0, 1\]


## Loss Functions

### Spatial Consistency Loss


### Exposure Control Loss


### Color Constancy Loss


### Color Correctness Loss


### Color Fine Tune Loss


### Illumination Smoothness Loss


## Result \(After Quantized\)