Could you please share which database the repertory had used? about blur-and-clear-classification HOT 6 CLOSED

Hi Jasonsey,
Very delightful that this repo is useful to you.
For the database part, it was captured by the security camera of my mentor for whole day motion.
As it is private so can't be shared.
But you can use it by feeding such similar pics.
Or you can find some videos and extract the frames out of it... But tagging part will take time.
As for this task I manually labeled the dataset and learned some good things ;)

Aditya

from blur-and-clear-classification.

Hi Aditya
Thank you for your answer. Maybe I can label some pics myself. And Could you please tell me what the standard used to label bad pics and good pics is.
For bad pics, the more blurred pic is better or the clear pic but with little blur is better.? And to good pics, the same.
I'm new to prepare such database. And Though it is a small hint, it is also important for my work.

Appreciate your help.

Thank you again.

Jasonsey

from blur-and-clear-classification.

Hi Jasonsey,
For mine task of labeling, I used the variance of Laplacian of Gaussian filter as an approximate technique to label the images, which helps to tag data faster.
You can search for it LoG filter.

Aditya

from blur-and-clear-classification.

Hi Aditya
It's very helpful to separate blurred pics from clear pics. But I have encountered a new problem. I have a database that contains many clear pictures, but fuzzy pictures are few. And My deep learning algorithm needs a lot of fuzzy pictures. So I want to create fuzzy pictures myself. Have you tried to generate blurred blurred pics from blurred pics?

Appreciate your help.

Thanks and regards.

Jasonsey

from blur-and-clear-classification.

from blur-and-clear-classification.

Closing this issue as no updates got

from blur-and-clear-classification.