mohitlamba94/restoring-extremely-dark-images-in-real-time

The project is the official implementation of our CVPR 2021 paper, "Restoring Extremely Dark Images in Real Time"

restoring-extremely-dark-images-in-real-time's Issues

test demo

Hello, may I ask which data set the model loaded in the demo is trained from, how big is the tested original image, and how generic the model is？

Variable ta is not defined in function part_init in common_classes.py of train_test_ours folder

Could you fix this problem?

HOW to use JPG OR RAW to run？

Whether to conduct data cheating in the experiment

1. sid and did use 0/* for training and you use 0/* +2/* for training?
2. sid and did use data config(0.1+0.04->10/30) and you use (0.1->10/30)?
   more training data and easier data complexity, is that fair????

You reply
MohitLamba94/LLPackNet#4 (comment)

1. testing issue

# from https://github.com/MohitLamba94/LLPackNet/blob/master/ablations/LLPackNet.ipynb
class get_data(Dataset):
    """Loads the Data."""
    
    def __init__(self,opt):
        self.train_files = glob.glob('/media/mohit/data/mohit/chen_dark_cvpr_18_dataset/Sony/short/1*_00_0.1s.ARW')
#         self.train_files = self.train_files + glob.glob('/media/mohit/data/mohit/chen_dark_cvpr_18_dataset/Sony/short/2*_00_0.1s.ARW')


        self.gt_files = []
        for x in self.train_files:
            self.gt_files =self.gt_files+ glob.glob('/media/mohit/data/mohit/chen_dark_cvpr_18_dataset/Sony/long/*'+x[-17:-12]+'*.ARW')
        
        self.to_tensor = torchvision.transforms.Compose([torchvision.transforms.ToTensor()])
        
        self.opt = opt

you used only 0.1s raw data for testing, why not use the below data for testing???

10003_00_0.04s.ARW  10030_00_0.04s.ARW  10045_00_0.04s.ARW  10178_00_0.04s.ARW  10193_00_0.04s.ARW  10217_00_0.04s.ARW
10006_00_0.04s.ARW  10032_00_0.04s.ARW  10054_00_0.04s.ARW  10185_00_0.04s.ARW  10198_00_0.04s.ARW  10226_00_0.04s.ARW
10011_00_0.04s.ARW  10034_00_0.04s.ARW  10055_00_0.04s.ARW  10187_00_0.04s.ARW  10199_00_0.04s.ARW  10227_00_0.04s.ARW
10016_00_0.04s.ARW  10035_00_0.04s.ARW  10068_00_0.04s.ARW  10191_00_0.04s.ARW  10203_00_0.04s.ARW  10228_00_0.04s.ARW
10022_00_0.04s.ARW  10040_00_0.04s.ARW  10069_00_0.04s.ARW  10192_00_0.04s.ARW  10213_00_0.04s.ARW
10178_00_0.033s.ARW  10187_00_0.033s.ARW  10192_00_0.033s.ARW  10198_00_0.033s.ARW  10203_00_0.033s.ARW  10217_00_0.033s.ARW  10227_00_0.033s.ARW
10185_00_0.033s.ARW  10191_00_0.033s.ARW  10193_00_0.033s.ARW  10199_00_0.033s.ARW  10213_00_0.033s.ARW  10226_00_0.033s.ARW  10228_00_0.033s.ARW

which was tested in sid and did.

# from https://github.com/parasmaharjan/DeepImageDenoising/blob/master/Pytorch_EDSR_Test.py#L120
in_files = glob.glob(input_dir + '%05d_00*.ARW' % test_id)
    for k in range(len(in_files)):

2. training issue

you said

For training I merge the 0* and 2* both of which belong to training phase

But for sid and did, they only use 0* data, totaly 161. BUT you use 181 data(more 20 data) for training?

# from https://github.com/cchen156/Learning-to-See-in-the-Dark/blob/master/train_Sony.py#L17
train_fns = glob.glob(gt_dir + '0*.ARW')
train_ids = [int(os.path.basename(train_fn)[0:5]) for train_fn in train_fns]

# from https://github.com/parasmaharjan/DeepImageDenoising/blob/master/Pytorch_EDSR.py#L39
train_fns = glob.glob(gt_dir + '0*.ARW')
train_ids = []
for i in range(len(train_fns)):
    _, train_fn = os.path.split(train_fns[i])
    train_ids.append(int(train_fn[0:5]))

Can you explain the above two issues?:)

js implementation

Thank you for the interesting solution.
It would be nice to have it ported to JS, lets say with TFJS.

ta i can't find ta

please response me,thanks!!!

Waiting for open source code

Hello

this is a test!

Ma metric

Hi, You did a really nice job, congratulations! Do you mind sharing how you get Ma metric? I am trying to use the code from https://github.com/chaoma99/sr-metric but it seems not very efficient. Is that the same way how you get Ma metric to put your test image directly to this process?

SID dataset training

Very comprehensive and brilliant work!But I have a problem when I try to retrain the Network in your project,I found that when you choose the train&test files in SID Sony Dataset, you just choose '0*_00_0.1s.ARW','1*_00_0.1s.ARW','2*_00_0.1s.ARW' and corresponding long exposure raw image as your dataset,which cause that you just use 181 pairs data for train and 50 pairs data for test,and ignore other exposure time low light image such as '*0.04s.ARW','*0.033s.ARW'

are you using same data config as sid or did

hi！I am trying to

Great work! Since your paper compare your method with LDC, I wonder where the code of LDC is.

LDC Learning to restore low-light images via decomposition-andenhancement CVPR2020

I cannot find its code. But you say that LDC [65] has publicly available code.

Can you point out its url or offer your reimplementation?

Thanks a lot!

Hello could you please provide the code just for amplification module，i'm a little confused this this part just reading paper,thanks

How to achieve parallelize in three scales？2x，8x，32x

These three scales 2x, 8x, and 32x in the network.py are executed sequentially in the forward of the same network, not in parallel

how can i test with jpeg or png

the test image format is raw, how can i test with the jpeg or png format

generalization of the amplifier module

Thanks a lot for your great work and shared code.
I tested your code on a Canon camera raw image and found that the amplifier module works poorly, while the camera final RGM image is very nice.

This is the output of your mode with external amp=1.0 (the outputs for amp = 2, 5 are way too over exposed)

This is the camera output

Do you have suggestions on how to improve the amplifier?

Training DCE model with SID dataset

I have a problem about how you train the DCE model with SID dataset ,did you postprocess the lowlight raw data with rawpy to get the RGB data and train the DCE model with those data?