1.11.3

Update mnnsr.
Add mnnsr for windows. It should support cpu/cuda/opencl/vulkan backends, but the cuda backend not works and without any warning on my computer.
Add mnnsr for linux ci, it shows how to build mnn and mnnsr with opencl/cuda/vulkan backends (the offical mnn prebuilt .so files only support cup, so build mnn is nessary) , but the ci build result mightly not works on your devices. I think you should build it by yourself
Add pth2mnn/onnx2mnn tools. you could use the web app or run it on your devices. https://huggingface.co/spaces/tumuyan2/model2mnn

1.11.2

1. fix some bug in mnnsr. fix some command error.
2. update mnnsr: display model info & ram usage (it is provide by mnn)
3. update mnnsr backend
4. update mnnsr: support different color space
5. fix model name for realcugan with custom param
6. add de-nearest2/de-nearest3 in resize-ncnn.
7. add 2 mnn model in assests.
8. show image resolution after loading image.
9. support avif image as input
10. update mnnsr: add decensor_mode (with -d 0 param). it could reduce mosaic with a normal super resolution model. if you have a onnx model train on nsfw datasets, it should be better.

1.11.1

1. feat: select multiple images at once (only preview 1 and the result will save to album automately). This feature have a toggle button to control. If this feature is disabled, you can still select multiple images from other apps (like Gallery) and share them to this app.
2. fix realsr-ncnn error time printing
3. fix bug: segmentation fault in realsr realcugan & waifu2x when you use cpu mode and select a image with alpha channel.
4. fix mnnsr-ncnn model path

1.11.0

1. Support mnn models. MNN framework could support more models than ncnn, but it is slower in my experiment. If you've tried converting models to ncnn format before, you must have been bothered by conversion failures. Using mnn can greatly reduce this problem. you could convert ONNX, TFLITE, TorchScript, Tensorflow models to mnn format.
2. Update realsr-ncnn, not require input/output layer names. If you convert other models into the ncnn format, you no longer need to manually edit the names of the input and output layers.
3. Update readme file in git repo. Add chainner description and provide a chainner project to convert model to ncnn format.

1.10.0

1. New feather: support gif animation input/output. It is not perfect, but he is practical.
   • Processing can only be done in GIF animation mode when one GIF is opened; otherwise, only one frame can be processed.
   • Only one frame can be previewed.
   • The result will be saved directly to album.
   • Magick (including the quick menu in the top right corner) does not support GIF processing.
2. Update ncnn/magick/opencv/libwebp

1.9.7

1. fix reset-assets
2. adjust realsr-ncnn log printing (limit progress update 2 times/sec, add build time info, print progress for cpu mode)
3. finetune realsr-ncnn tilesize for Snapdragon (This adjustment only affects Adreno GPU when you use a big model.finetune according to my 8gen2 device. )

1.9.6

1. Add model RealeSR-general-v3 , a official small model, fast but not very good.
2. Upgrade gradle and jnilibs (libncnn, magick, stb)
3. Release latest assets package
4. Upload a onnx2ncnn tool. You can use the mouse to drag an ONNX model to onnx2ncnn.bat to convert its format, without manually input commands or modify the param file. However, since the program is unable to determine the scaling of your model, you still need to manually rename the files.

1.9.5

1. Fix #56 by upgrade magick. https://github.com/MolotovCherry/Android-ImageMagick7 has been archived, now I use this fork https://github.com/Miseryset/Android-ImageMagick7
2. Add menu: reset cache
3. Update readme

1.9.4

1. Reduce save time (opencv instand of stb lib), it could be userful for process big image.
2. Update help information (add anime4kcpp link)
3. Fix de-nearest. (Although it is not useful.)
4. Update Magick to 7.1.24. ( https://github.com/MolotovCherry/Android-ImageMagick7 has been archived by the owner on Jan 25, 2024, many thanks to him)

1.9.3

1. Remove srmd models, I think it's useless. Anyway, if you need it, you could download it and apply it as extra models.
2. Add real-esrgan model Nomos8kSC, it was trained by Phhofm, a nice model for photos.
3. Change models reop in readme, now the extra models is in https://huggingface.co/tumuyan2/realsr-models
4. Add interpolation config options, you could remove any interpolation command or add others magick filters easily. I think you could read this: https://imagemagick.org/Usage/filter/