File size: 2,421 Bytes
00c3521
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
# Team: **play1play**

for 'Night Photography Rendering Challenge'

This repo contains the source code of [Night Photography Rendering Challenge 2024](https://nightimaging.org/).

## How to run
### Run without Docker
- Install python >= 3.9
- Install the required packages: `pip install -r requirements.txt`
- Put the test images in the `data` folder or specify the input image path with `-p` option.
- Run the script: 
```
python run.py -p <input_image_path> -o <output_image_path>
```
- The output images will be saved in the `output` folder or specifiy the output path with `-o` option.

### Run with Docker
- Build the docker image from beginning (optional): 
```
docker build -t play1play .
```
- Run the docker container with gpu on linux:
```
docker run -it --rm --gpus=all -v $(pwd)/data:/data play1play ./run.sh
```
`Do not forget to to --gpus flag, our model requires GPU to run.`

## Update
*2024.3.21:*

Final version v3.0 is released for 3rd validation!

Key features:
- Utilize the patch-based and calibration-based white-balance algorithm to improve the image quality.
- Modify the resizing strategy to improve the image quality.

*2024.3.16:*

Release v2.0 for 3rd validation!

Key features:
- Increase the overall saturation and brightness of the image.
- Add more contrast to the image.
- Add more dynamic range to the image.

Algorithm changes:
- Add luma shading correction (LSC) module
- Add auto-contrast module (dynamic gamma)
- LSC, LTM, auto-contrast module can dynamiclly adjust the parameters based on the camera gain from the metadata.
- Add another white-balance process at the end of the pipeline to further improve the image quality.

Key algorithm parameter explanation:
- k_th: defines the threshold for the noise level, higher means tolerant to noise, lower means more sensitive to noise. For this sensor, default is 2.5e-3. `Note that this paramerts are shared by all the modules, and hence defined as the member varibale in RawProcessingPipelineDemo class.`
- s of local_tone_mapping: defines how to apply the gain map to different image channels. Higher means more saturated, lower means less saturated. Default is 0.7. `Currently, s is automatically adjusted based on the camera gain from the metadata.`

*2024.3.7:*

release v1.0 for 3rd validation!

----

## Version-1

- TMO-ratio50

-----

## Version-2

- TMO: ratio50

- Gamma: 1.5

- Contrast:[low=2,high=0.2]

- Post-AWB: GI