Skip to content

shitian-ni/CUDA-GPT

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

58 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Accelerated GPT in CUDA

Usage

Compile and Execute

nvcc *.cpp include/*.c* -o cu.out  -arch=sm_60 --use_fast_math

Modify include/parameter.h Line 142 from #define MAKETEMP 0to #define MAKETEMP 1 for table generations. Compile and execute. Then change back to #define MAKETEMP 0.

  • If the step above is not done then Cannot open the file! error will arise.

Tested on Tokyo Tech TSUBAME 3.0

  • Nvidia Tesla P100
  • Cuda compilation tools, release 8.0, V8.0.61

Known issues

minor result difference between CPU and GPU

  • Could be CPU precision error in multiplyVect3x3 used in bilinear_normal_projection function in stdGpt.cpp

Reference

    @inproceedings{
        title={Theoretical criterion for image matching using GPT correlation},
        author={Shizhi Zhang, Toru Wakahara, Yukihiko Yamashita},
        booktitle={2016 23rd International Conference on Pattern Recognition (ICPR)},
        year={2016},
        DOI={10.1109/ICPR.2016.7899692}
    }
    @inproceedings{
        title={Image Matching Using GPT Correlation Associated with Simplified HOG Patterns},
        author={Shizhi Zhang, Toru Wakahara, Yukihiko Yamashita},
        booktitle={2017 7th International Conference on Image Processing Theory, Tools and Applications (IPTA)},
        year={2017},
    }

License

Apache License, Version 2.0