Open joel99 opened 5 years ago
Something I didn't investigate that much is the input and training style size. If I recall correctly, the current model gives different results depending on the resolution of style image used for training. It would be nice to tune this to whatever size input image we use. Also I think the input image size was hardcoded based on a scaling of the default resolution of the camera we were using.
Mainly https://github.com/xunhuang1995/AdaIN-style. Segmentation in photobooth remains the same, but for an artistic endeavor the possibilities in photobooth should be expanded for more sophisticated techniques.