meijianhan / deepdeblur Goto Github PK

I don't know if it relevant but I am running your code on windows 10 with Keras version 2.7.0 , TensorFlow version 3.1.1 and python 3.8 (64 bit) and every time I get the error "cannot pickle 'generator' object". Now I know that this problem arises because pickling a generator is not possible with current versions of python as it poses security risks. Instead I am trying to resolve this issue by replacing generator expression by list() expression but i have not been successful in doing that yet. So my question to you is did you face this issue when you were running it with Ubuntu.
If no then another of my question to you is that do you think it has got something to do with the the OS or the versions of library functions.
If yes how did you get rid of that?

And I apologise if this entire issue is a very vague question.

I was able to solve this with a workaround for now by making the multiprocessing to False flag and then running the code but it takes Herculean time so i had to change the epoch values and steps in every epoch value to a really miniscule level.

Any advice from you is much appreciated

Hi, I am trying to train the model with my own data, I have images of varying dimensions when I am trying to run it this error is popping up

PATH
path_sharp = '/home/jediyoda/archit/Dat_Noh/train/train_h/'
path_blur = '/home/jediyoda/archit/Dat_Noh/train/train_l/'
Params
batch_size = 128
epochs = 21
input image dimensions
img_rows = 65
img_cols = 65
num_of_dim = 3

Using the following parameters to try training the model.

Hi. @meijianhan The code is used for any paper? Or not?

Hi! Does this code have a corresponding paper?
Thank you!

is it necessary to have 65X65X3? I ran your MTABLAB function on my images and the dimension was 65X65. I was able to train the model. But when I test I face the following issue:
ValueError: Layer #1 (named "model_1"), weight <tf.Variable 'conv2d_19/kernel:0' shape=(3, 3, 128, 1) dtype=float32_ref> has shape (3, 3, 128, 1), but the saved weight has shape (128, 128, 3, 3).