Skip to content

Home

Jump to bottom
Raphael Kim edited this page Aug 15, 2018 · 11 revisions

libsrcnn

Features

  • Native C++ source codes for G++ and MinGW-W64.
  • No OpenCV needed.
  • Much faster within 4x ( depends on Cores of CPU ).
  • OpenMP accelerated.
  • Suitabled for almost of platforms - ARM, ARM Cortex-A, AARCH64, x86.32, x86.64
  • Less memory than OpenCV model.
  • More interpolation filterings - Bicubic, Lanzcos-3, B-Spline.
  • Easy to use
  • LGPLv3

Tested systems

  • Intel Core2Duo, Dualcore x86.64, Windows 7 64bit and 10 64bit w/ MinGW-W64
  • Intel Core i5 2450M, Windows 10 64bit w/ MinGW-W64
  • Intel Core i5 3570, Windows 10 64bit w/ MinGW-W64
  • Intel Core i5 7200U, Windows 10 64bit w/ MinGW-w64
  • Raspberry Pi 3 Model B+ w/ built-in gcc ( armhf )
  • nVidia Jetson TX2 w/ built-in gcc ( aarch64 )
  • BeagleBone Black w/ built-in gcc ( armhf )
  • Snapdragon 435 ( EMUI 5.1, Android 7.0 ) Termux GCC 6.0.1 ( aarch64 )

Not supported

  • M$VC - no plan to support this junk.

Issue

  • No OpenMP for MacOSX, you can use HPC-GCC but it should bring more compatibility with Apple compiler.
  • My project supports LLVM-GCC ( clang symlink ) to building library and testing programs.
  • Performance at MacOSX may not wonderful, took 6 sec for 512x341 RGB image scaling up to double - hope Apple can support OpenMP in the future.
  • M$VC stopped to support higher version of OpenMP, instead of, they are using concurrency - I don't have any plan to use their sucking whatever.
Clone this wiki locally