vavi-image-filter-dnnsuperres

_{© OpenCV}

BufferedImageOp using pretrained models to OpenCV's DNNSuperRes

this is a fork of https://github.com/Araxeus/PNG-Upscale

Install

• maven

 $ mvn -P lwjgl-natives-macos-amd64 -Djavacpp.platform=macosx-x86_64 install

Usage

    DNNSuperResolutionOp filter = new DNNSuperResolutionOp(DNNSuperResolutionOp.MODES[0]);
    BufferedImage image = ImageIO.read(Files.newInputStream(in));
    BufferedImage filteredImage = filter.filter(image, null);
    ImageIO.write(filteredImage, "PNG", Files.newOutputStream(out));

TODO

• quality
  • models this project provided ...
    • sample image makes good result, but others...
    • otoh CoreML REAL-ESRGAN creates significant result (color is somehow different...)
  • https://github.com/umjammer/vavi-image-filter-ml/blob/main/src/test/resources/namacha.jpg ... i couldn't see differences
  • https://qiita.com/gomamitu/items/b4722741f6318d734bce ... i think there is no differences (magnifying is needed lol)
  • https://meknowledge.jpn.org/2021/05/28/python-super-resolution/ ... same impression
• Dnn#readNetFromTensorflow()
• resources in "jar in jar" for mac .app
  • dnnsuperres error in mac.app
  • because not "resources in jar (we can see those as individual files)" but "resources in a resource jar (we can see the jar only)"
    • -> need to extract