Qengineering / realsr-ncnn-Jetson-Nano

Award winning RealSR super resolution with ncnn on Jetson Nano
https://qengineering.eu/deep-learning-examples-on-raspberry-32-64-os.html
BSD 3-Clause "New" or "Revised" License
36 stars 11 forks source link

Request for deblurring application #3

Closed rsingh2083 closed 3 years ago

rsingh2083 commented 3 years ago

Dear Sir,

By any chance can you develop a fast and lightweight image de blurring application ? Just a polite request.

Qengineering commented 3 years ago

That's not so easy to do either. The most common way is to use an FFT kernel that describes the desired blur. You need one 2D forward FFT, a convolution phase and a backward FFT. FFTs are usually 2^x sizes (32,64,128 etc.). Other formats requires additional algorithms. As you can imagine, a lightweight solution is not easy to find. Often, the results are also disappointing. Only very well defined blur can be reasonable eliminated.

rsingh2083 commented 3 years ago

Thanks for the suggestions sir