README.md

Using deblurv2 GAN on custom dataset

Dataset preparation

You can download any set of images which are text, scenery etc . Using gaussian blur you can blur a copy of the images. I have took a dataset from roboflow . Made following folders

--Licenseplate_blur_clear_dataset
    --train
        --blur
        --sharp
    --test
        --blur
        --sharp
    --valid
        --blur
        --sharp

Copy original images of train,test,valid to sharp folder. To fill the images in blur folder you can use this code. now we are done with data preparation. Dataset which i prepared can be found here

Installation

pip install -r requirements.txt

Train

python train.py

Test

pretrained weights can be found here

python predict.py -i ./test_img -o ./submit -m ./saved_models/best_fpn.h5

Gallery

Colab Notebook

Notebook

Credits

• DeblurGANv2