505030475 / extremelowlight

Code&Dataset : Learning an Adaptive Model for Extreme Low-light Raw Image Processing

ExtremeLowLight

Code&Dataset : Learning an Adaptive Model for Extreme Low-light Raw Image Processing

DATASET

CID (Campus Image Dataset) is a dataset capture in low-light env with the help of Android programming. Its basic unit is group, which is named by capture time and contains 8 expoure-time-varying raw image shot in a burst.

e.g.

Capture Program

Run on android platform, on some old device it seems that it has compatibility problem, please build from source.

source: /Camera Raw Burst/*

apk: /Application-debug.apk

This program is based on Google Camera2RAW project, and developed in a very limited time to serve very specific purpose. Thus, the UI is not well designed, you might need more time to understand it.

Android SDK required:Android 8 Oreo API Level 26

    minSdkVersion 21
    
    targetSdkVersion 26

Gradle required:gradle-4.4 (https:/services.gradle.org/distributions/gradle-4.4-all.zip)

usage:

(1) Press the Make Folder button to make a working dir in /storage/emulated/0/000dateset/. All images captured will be put into 000dateset folder.

(2) Adjust 3 seekbars (ISO, shortest exposure time, longest exposure time.

(3) Fix the device, press the BIG burst button to commit burst. Now the program will play a music as a reminder that you shoud HAND OFF the cam, and the program will sleep for a few seconds, waiting for the vibration of tripod to decrease. Meanwhile the screen no longer update, because the burst must be finished as fast as possible and screen update will interrupt this process.

(4) Press burst again or exit and find the results in /storage/emulated/0/000dateset/

DATASET File Organization

1、For trainning, provided as group

2、For evluation, provided as group or single

DATASET Download

OneDrive : https://stuhiteducn-my.sharepoint.com/:f:/g/personal/1150700117_stu_hit_edu_cn/EumyrYnHDMtIiNoAHNshbXUBHU5diRWyGVB1rmso860kog?e=KKEKOO

Pretrained Model Download

OneDrive : https://stuhiteducn-my.sharepoint.com/:u:/g/personal/1150700117_stu_hit_edu_cn/Edor2A427PtElKU6Y0ewvysBv0t3k0UM43mE2xxDFGenlQ?e=VqEyhG

extremelowlight's Issues

White image results

I tried running the python code on several DNGs obtained from iphone camera but all of them results in blank white images:

Here is a sample DNG:
https://drive.google.com/file/d/1G2T2wowpEcbgnqtzr-9Ir4hQ82ON4IfL/view?usp=sharing

Environment details:
Python: 3.8
Tensorflow: 1.12
opencv-python: 4.4.0.42

What am i doing wrong?

test our data

Hello, I test jpg picture find error,but test dng picture will not go wrong, excuse me how to convert jpg format picture into dng format

Downloaded dataset corrupted after unzipping

After I unzipped the dataset downloaded from the given link, some data are missing and reporting an error, and data in CID dataset.Z01, .Z02, ..., .Z06 and .zip are nearly the same. So I'm a little confused. Could you please have a look into this data issue?

Too Many Bugs on Your dataCapture code

I try to use your datacapture code on my google pixel2; but too many bugs flood me;

1. values missing on your res folds; I fixed it;
2. bad sync between cameraCapture and ImageReader;
   if(tor ==null) sleep(69) can't work.
3. counter_down in class ImageSaverX does't work
4. are you really from HIt?