Ezsynth

Small bug fixes--
Single Style image now works properly.
Still no luck with linux.

What's Changed

• slight Refactorings · 0d0b6d9e by @Trentonom0r3 in https://github.com/Trentonom0r3/Ezsynth/pull/27

Full Changelog: https://github.com/Trentonom0r3/Ezsynth/compare/2.0.3...2.1.0

Significant Refactoring of the Library. (In My opinion, easier to follow).
Various Refactorings, separations, etc to classes.
Utilization of opencv for warping vs using torch.
Improved computation time, down to ~4 minutes. (Still not great, but much better than previously ~10 minutes).
Implemented multithreading in Ezsynth, ImageSynth remains the same.
Added logic for usage of .so with Linux. If someone can build and contribute the .so, would be much appreciated.
Went back to .dll usage over .pyd.
Perhaps it was how I was setting up the .pyd, perhaps its was something else, but using the .dll and utilizing multithreading leads to huge performance gains I can't ignore, so I scrapped the .pyd.
Upload new version to pypi.

Minor Update. Passing outputpath to ezsynth.run results in not only the final images being saved, but the inbetween images as well.

from ezsynth import Ezsynth


STYLE_PATHS = [
    "output000.jpg",
    "output099.jpg",
]

IMAGE_FOLDER = "C:/Input"
OUTPUT_FOLDER = "C:/Output"

ez = Ezsynth(styles=STYLE_PATHS, imgsequence=IMAGE_FOLDER, flow_model='sintel')

ez.set_guides().stylize(output_path=OUTPUT_FOLDER) # Saves in-between imgs AND final imgs.

# results = ez.set_guides().stylize() # returns a list of images as numpy arrays

Updated to pybind11 wrapper, ImageSynth Class, and new Ezsynth class for video stylization!

Few small updates, changed backend to auto (you can still explicitly set backend if you'd like), changed votemode to weighted for sharper results.
Added some extra comments to give better detail.

Few small updates since initial commit.
Use file paths and/or numpy arrays.

Please use the most updated version [1.0.1]

pip install --upgrade ezsynth

or

pip install ezsynth