bookerdewitt / msbdn-dff Goto Github PK

Hi,
I want to take a look at what's in model.pkl. However, when I open it using the following code, the output is a series of number.

import pickle
info = pickle.load(open('./model/MSBDNet/checkpoint/msbdn_model.pkl', 'rb'), encoding='utf-8')
print(info)

Output:
119547037146038801333356

Is that the decoding problem? Or do I need some specific package to unpack it, if so，could you please tell which one should I use?
Thank you very much.

Thank you for your excellent job about the haze removal task. Why don't use the skip-connection between the input hazy image and the output image? I am a little puzzled about it. Is it conflict with the novel design of the network?

"Download datasets from the provided links and place them in this directory"
I want to know where is the links?

Hi, Thanks for your good jog for image dehazing. And also for the sharing of all the code and models.
I want to do a comparision of the detection results on the dehazed image , can you give a code link for yolov3 for fair comparsion?

Why is the result of training fuzzy when using your own data set?

If you release the training code, I'll appreciate it

图片大小：6464*4464均为16的倍数，但仍然报错

HI,
I've just started learning machine learning for a few months，somethings I don't understand.
I want to know in class "ResidualBlock",out= self.conv2(out) * 0.1, why 0.1?

您好，我在Epoch5之后出现了如下错误，请问要怎么解决呢：
Traceback (most recent call last):
File "train.py", line 163, in
opt.nEpochs = training_settings[i - 1]['nEpochs']
IndexError: list index out of range

How can i input a hazed_image then output an image result

File "train.py",line 183,in
with open("Logs/Output_{}.txt".format(opt.name), "a+")as text_file:
FileNotFoundError:No such file or directory: 'Logs/Output_MSBDN-DFF.txt'

Hello, how can I solve this problem?

Training data set details. The.h5 file you provided does not know which data sets the model was trained on.

After tuning the code, the result can only be run on one video card

Please！

excuse me , I have a question.
when I test with my date set , it comes the error : The size of tensor a (1088) must match the size of tensor b (1080) at non-singleton dimension 2 .
My data set end with '.png' , I had try the other images which end with 'png',but still meet this error.
But the image end with ‘jpg’ works well.
how can I solve it. thanks a lot.

Where is the code to generate the .h5 file?

python test.py

RuntimeError: The size of tensor a (624) must match the size of tensor b (620) at non-singleton dimension 3

How does this change from MSBDN-DFF

Thank you for sharing your great code. 😺

What is the license for this model? I'd like to cite it to the repository I'm working on if possible, but I want to post the license correctly.
https://github.com/PINTO0309/PINTO_model_zoo

Thank you.

Please！

Hello, I want to change a data set for training, but I found that the data set format used in this paper is. H5, I don't have data set in this format. I only have data set in. JPG format. How can I modify the training code so as to use my own data set? Or do you have any specific methods to change my data set format to. H5 format? This data set in. H5 format is very special, and there must be a lot of conversion details when you synthesize it. Please let me know Thank you very much

FileNotFoundError: [Errno 2] No such file or directory: 'Logs/Output_MSBDN-DFF.txt'

Hi,thanks for your sharing!
I have read your paper and i am very interested in the dataset called KITTI Haze dataset.
I will very appreciate if you provide the download link of this dataset.

Best Wishes,

How to load model.pkl file?

python test.py --checkpoint model.pkl command failed when load model.pkl

model = torch.load(join(opt.checkpoint, test_list[i]), map_location=lambda storage, loc: storage)

Hi
Thank you for sharing your research and open-source code. I was able to replicate your results using the pre-trained weights. I wanted to train a model, for which I require the training set.
In the paper, you have mentioned

To learn a general dehazing model for both indoor and outdoor scenes, we select as the training set 9000 outdoor hazy/clean image pairs and 7000 indoor pairs from the RESIDE training dataset [33] by removing redundant images from the same scenes.

Could you provide those particular images.
Thanks

I would like to know if the author has cropped the high resolution images, I use 2080TI can not directly handle these images, it will cause the video memory overflow.

What is the role of GRes? If I remove this module, what effect does it have on the result?

I want to know whether changing the data into. H5 file will affect the effect of network training? I used my own data set for training, and I didn't convert the image into. H5 file, but after preprocessing the data, I input it into the network. The network uses the network you proposed, but why does the output result have mosaic effect? The effect is very bad?

When I run test.py using the model.pkl file provided by the author, I ran into a problem:
ModuleNotFoundError: No module named 'networks.MSBDN-DFF-v1-1'
Does anyone encounter this problem and how to solve it？

Excuse me, but can you explain why you didn't use any batch normalization layer in your network ?

Excuse me，How to solve the problem " AttributeError: 'UpsampleConvLayer' object has no attribute 'reflection_pad'"
Thank you!

hello,how to deal with "local variable 'iteration' referenced before assignment"?