mahmoudnafifi / exposure_correction Goto Github PK

hello , i am really interested in your paper and look forward to your open source in the later ,thank you

hey @mahmoudnafifi

This is a wonderful work, but since the implementation is in MATlab the reach is limited.
I was thinking of reimplementing this research paper in python, before doing that I thought first check with you if you are fine with it or not. I also dropped a mail few weeks back.

So will that be fine with you ?

I will definitely put citation, refence to original repo and research paper

Hi, I am looking to download the MSEC dataset proposed in the paper. However, the links are invalid. Could you please share the updated link for the train, validation and test?

Thank you in advance.

https://drive.google.com/file/d/1A3p_E0ajR22ZNyPw_gUlaQN8RlBvQIcX/view

When are you going to release the dataset? -Thanks

Hello,
i'm a student and i read your paper, i'm trying to create the dataset and i would like to ask you if it is possible to have the list of images of the test set.
Thanks!

I trained the model with the original paired images of ExpertC in MIT-Adobe-5K, only underexposed and well-exposed images are provided.
When I run the demo, I got almost the same image as input, what happend ?

Left is the input and the right one is corresponding output using model_512.mat

Hello, could you please put the dataset on Google Drive? Downloading from 'sync.com' is really slow.

Thank you for your excellent work! Will the code be released soon?

When are you going to release the dataset? -Thanks

Hi @mahmoudnafifi

Thanks for the excellent work on exposure correction!

I tried to retrain your code using my data. After training, I obtained three generator models and three discriminator models.
I wonder which model can be used to generate the final results. I have tried to replace the 'model' in the demo_image_directory.m with 'model_128', 'model_256', and 'model_512'. But I cannot obtain the correct results.

Could you please give me some advice to generate the correct results using the retrained models? Thanks！

Hi, @mahmoudnafifi

While code for this project is not released I'd like to clarify which implementation of bilateral guided upsample did you use?
I found 3 versions:

1. Code from Google google/bgu.
2. Code from Google's HRNet paper google/hdrnet
3. Version from DeepUPE repo Jia-Research-Lab/DeepUPE.

All these options have their own limitations, 1 and 3 is pretty old and can't be compiled with new CUDA10+ and TF2+, while 2 is GPU only.

So my questions is: have you written your own custom implementation or adapted one of those?

Hi, @mahmoudnafifi!

Could you please share your loss curve during training? For both, disc and generator

Thanks in advance!

hello,would you want to share the python pytorch version's code?hahh

My GUI doesn't fit in the window (it significantly goes outside the window, including the Browse button outside the window). Apparently because my screen scale factor in Windows is 200%. Is it possible to make some changes in demo_GUI.m code (or maybe in Matlab IDE setting) not to change current screen scale which suits all other applications (and my eyesight :) ) ?

Hello Dear Author.

I have noticed that most of the images in the validation and test set are not exactly powers of 2, Therefore,
when downsampling them in Unets like architectures the dimensions won't exactly match in the encoder and decoder when doing skip connections. This is not an issue when it comes to training since patches can be extracted. However, when evaluating or testing on full-size images from the dataset you proposed it becomes an issue.

My questions are the following :

1. How do you resize them (the images) before performing your evaluation and testing?
2. Or, do you extract patches from them? if so what size of the patch and what location?

Thank you .

Hope to see the released code asap.

Hello, I observed that the images in your dataset do not have metals, can I use the dataset and the network if the metal is overexposed and some information is missing in the images?

I noticed the code is based on Matlab, which is very expensive...
It's greatly appreciated if a Pytorch(free to use) version of the code provide!

Hello, @mahmoudnafifi ! Great work, thank you for that.
Do you use pyramid loss in this matlab implementation ? Because I haven`t noticed

I tried to modify the code, but failed to create the generator.