Skip to content

jeeneo/dejpeg

Repository files navigation


an app for removing noise and compression from photos

an image of two rectangular areas of the same picture of yellow tulips, the one at the top is 50 percent compression, the one below has 0 percent compression but still has noticeable PNG compression artifacts an image fruit, two purple figs sitting directly on a table. one fig is behind the other fig. the fig in the front is cut in half but still remaining upright. text on the left near the first fig reads fig 1. text near the fig on the right reads fig 2. both figs are sitting on top of a table that has the text table 1. there's a slider down the middle of the image showing the differences between a processed image and the original one

IzzyOnDroid Obtanium Get it on GitHub

features:

  • batch processing
  • supports most image formats
  • before/after view
  • fully offline

(this is not a "super resolution AI upscaler", but a simple non-destructive way to clean up compressed/noisy images)

models (required):

FBCNN (JPEG compression)

SCUNet (Grain/Noise)

info here (also in the apps FAQ)

you can also run other experimental models, more info here

limitations:

  • processed locally, minimum 4gb ram and 4 threads recommended

see the wiki for more information

desktop

for desktops, you can try chaiNNer, it has a bit of a learning curve and takes a bit to set up, but works pretty well and has some decent GPU acceleration.

note:

DeJPEG is not affiliated or related with Topaz DEJPEG or any other similarly named software/project

credits:

@adrianerrea for base application, FBCNN and SCUNet creators and all other model creators, this is only a wrapper for 1x ONNX models

license:

all models used are under their respective licenses

you are free to embed parts of this app in your own project as long as it remains free/non-paywalled and must abide to the GPL v3 license