README.md

MobedSR: Practice of Super Resolution on the Android

TODO

• Implement the frame of app (23/01/25)
• Implement TF Lite Interpreter by Java (23/01/25)
• Implement TF Lite Interpreter by C++

• Build DNN model using the python a to z
• Add evaluation indicators i.e. PSNR, SSIM
• Expanding the subject from Image Super Resolution to other applications
• Implement the CPU, GPU co-execution for DNN task (GOAL!)

• Implement Video Super Resolution by using FFmpeg
• Refactor the VSR code more generally
• Solve the issue of storage -> it's hardcoded now

Progress log

23/01/25:

• Start Toy Project.
• Finish setting up basic frame of app.
• Finish java implementation of simple super resolution app

23/01/26:

• Expand the project to video super resolution
• Create new activity for video super resolutio
• Capture frames from video by using FFmpeg

23/01/27:

• Implement split the video by fps
• Ongoing on super resolution the frames and convert them to the super resolution video

References

• https://github.com/arthenica/ffmpeg-kit/tree/main/android
• https://github.com/cd-athena/MoViDNN
• https://openaccess.thecvf.com/content/CVPR2021W/MAI/papers/Liu_EVSRNet_Efficient_Video_Super-Resolution_With_Neural_Architecture_Search_CVPRW_2021_paper.pdf

Low resolution video

https://commons.wikimedia.org/wiki/Category:Video_display_resolution_480_x_270