misaleh / libelas-gpu

Implementation of LIBELAS in cuda. with opencv APIs

libelas-gpu

This is an implementation of the well known libelas (LIBrary for Efficient LArge-scale Stereo matching) library for sparse large real-time calculation of stereo disparity images. This is for a college course final project and thus does not have any support of any kind. The original source code can be found in the CPU directory. This has been commented as we explore the code.

From there, the GPU cuda implementation of key methods can be found in the GPU folder. For methods not implemented on the GPU the CPU version is called, and such code is run on the CPU.

Differences

The key difference is the changing of most of the methods in the ELAS class to being virtual. This allows for the sub-classes to override such methods. This allows for seamless transition between the GPU and CPU code. Additionally, the methods where converted to public to allow for the testing and comparison of the two methods directly.

Building

• First create a build directory mkdir build
• Move into this directory cd build
• Run the c++ cmake build system cmake ..
• Finally build all the executables make -j5
• To build a single one preform the following make -j5 <name-of-exc>
• To run the program do the following
  • ./libelas_cpu demo
  • ./libelas_gpu

In this fork

• This fork focuses mainly on improvement of GPU implementaion of the algorithm.

• Added opencv mat support with GPU (can be ported easliy to CPU), so you can read, wirte matrices and make further processing using opencv APIs.

• Changed default parameters to MIDDLEBURY as it produces better output.

• Implemented leftRightConsistencyCheck on GPU

• Implemented Sobel filter on GPU

• Implemented Median filter on GPU

• Edited the code to work without SSE so it can run on arm processors

• To build the GPU code with the new changes just type make in the terminal

• To run type make run

Licenses

All the original code is licensed under the GNU General Public License. This can be found with the original libelas library. All other code is licensed under the MIT license, which is attached to this repo's LICENSE file.