jinyeying / fogremoval Goto Github PK

[ACCV22] Structure Representation Network and Uncertainty Feedback Learning for Dense Non-Uniform Fog Removal, https://arxiv.org/abs/2210.03061

Python 97.15% MATLAB 2.85%

deep-learning defog defogging dehaze dehazing smoke dense

fogremoval's Introduction

Hi 👋 I'm Jin Yeying

I am a Senior Researcher at Tencent. I am pursuing my Ph.D. degree at the National University of Singapore (NUS), supervised by Prof. Robby T. Tan. I had my research internship in Adobe, mentored by Prof. Connelly Barnes.

Previously, I completed my M.Sc. degree from the National University of Singapore (NUS); and received my B.Eng degree from University of Electronic Science and Technology of China (UESTC).

My primary research interests include computer vision and deep learning, mainly focusing on image/video generation and enhancement.

Languages and Tools:

fogremoval's People

Stargazers

Watchers

Forkers

viml-cvdl ip-dehaze islandlz paperwave rohanhbtu moonysunny2

fogremoval's Issues

Comparsioins

Can you give other results of comparsions, such as MSBDN, PSD and Dehamer.

Asking for help about training code

I am interested in your paper, but we meet some difficulties during implementing your algorithm. Do you have any plans to release the training code?

comparsions

Did you train the network of comparsions on the smoke dataset? but you train your network on the smoke dataset. Whether such a comparative experiment is convincing?

Dataset mirror

Hi,
Thank you for your work. Unfortunately, the dropbox link look broken to me.
Is there another way to get the data?

Thanks in advance!

Training Setting Question

Thanks for the nice work.

I'm confuse about the training setting in paper, which dataset have you use to first "pre-train" the model?

And then use the "pre-train" weights to fine-tune on various dehazing dataset (like the smoke dataset)? Am I understand right ?

Hope to get your reply! Thx ~

Hi! First of all, thanks for your work, it seems really interesting to me. I'm trying to test your model on some images of SMOKE Dataset loading "checkpoint.1000.ckpt", but it appears to be empty. Would it be possible to update the weights or download them another way to test the model?

Thanks in advance.

O-haze

output image is strange

Sorry to bother. I use your code to work on my own real-world image and found my output is very strange, but when I use test images in directory provided it shows normal. So can you tell me how to deal with it ?
My output image url : https://github.com/2bhuanleduo/git-demo/blob/master/2_25.png

Uncertainty Map

The generation illustration of Uncertainty Map is unclear in paper. More specifically, I cannot understand the paramater "theta". The paper just told it is the variance of the Laplace distribution without more information. If I want to train, how to get "theta"?
Can you provide more details?

"O Haze" dataset evaluation

Dear esteemed author, I would like to inquire about two matters:

It appears that pre-trained models for the "O Haze" dataset have not been provided. Could you confirm if all datasets were tested using the "NH Haze" pre-trained model you supplied?
The resolution of the dehazed images from the O Haze test set that you showcased is notably lower than that of the original dataset. Did you perform resizing on the original images during the testing phase?