fqhd/inpainting

A generative adversarial network that can regenerate parts of images.

Inpainting

A generative adversarial network that can repaint parts of images. If you want to learn about how this works checkout my youtube video on the topic. I explain in detail how the algorithm works, and ways in which you can develop and further improve the model if you wish to achieve state of the art results.

Dataset

The images of flowers that were used to train this model can be downloaded from here The bezier curves that are used as strokes can be downloaded from here downloaded as well.

Testing

There are 2 pretrained models saved under the models folder. If you wish to test these models, make sure you are in the root directory of the project and run python patch_painting/test.py to test the patch painting model or python stroke_painting/test.py to test the stroke painting model. Make sure you have strokes.npy downloaded for this.

Training

To train the model, make sure you have the Dataset and the strokes.npy downloaded. You also have to create a folder in the root directory of the project called progress. Making sure you are inside the root directory of the project, run python patch_painting/train.py to train the patch painting model, or python stroke_painting.py to train the stroke painting model.

Preview