PEN-Net has been written into Keras now! which means it is easier to use !
"PEN-Net" is CVPR-2019 paper for Image Inpainting task, see reference for more details!
PEN-Net Decoding process visualization:
Notice: This model havs some unknown issue which need to be fixed as PEN-Net(Keras) can not produce high quality visual realistic details, I am trying to fix it currently. If you know what's the problem, feel free to do the pull request!
-
scipy==1.1.0
-
pillow
-
numpy
-
tensorflow-gpu
-
keras
-
matplotlib
-
(programed in python 3.6)
- put correct model under the "/models" folder
- init the PENNet class and call the class function "test_app_console" for testing
- The purely white pix of the input img will be classify as the "mask" automatically
- the output image will be in the "/generated_Imgs/test_app"
- init the PENNet class and the relevant configs (batch_size,dataset_path,etc.)
- select a proper type(or way) of how dataloader works
- just do the training by running the program! The model and epcho of saving will be saved directly by PENNet class automatically
Licensed under an MIT license.