Training the Inpainting model

[waleedrazakhan92]

Hello, I've tested the inpaining model with some brush strokes of my own. But the results are not very good. Through experimenting I've figured out that the width of the brush strokes that I'm using is the difference between getting good vs bad results.
Moreover I've also come to have known through testing that the quality of results is quite susceptible to the pattern of the brush strokes. The zig zag strokes used in the GPEN is quite different from my circular brush strokes which are also more roundish.

So I'm planning to finetune/train the inpainting model for my problem. So, did anyone trained the image inpainting model successfully with comparable results to the paper? It would be great if someone can share some helper code and the reference dataset they used. Basically anything to kick start the training. I've gone through the issues but didn't find any leads about the inpaiting training.