search

CTU-IIG / kcf

Kernelized Correlation Filter tracker
13 stars 6 forks source link

Lessblocks #52

Closed FlavioK closed 3 years ago

FlavioK commented 5 years ago

Kernels have been changed to run with only two CUDA-blocks à 512 threads. Gaussian correlation and CublasScale are now running as normal kernels on the GPU. sqr_norm kernel has been rewritten to use __shfl_down_sync, see: https://devblogs.nvidia.com/faster-parallel-reductions-kepler/ A lot of memcpy operations to and from the device have been removed.

FlavioK commented 3 years ago

I close this pull-request since its not needed anymore. I hope this is ok. :-)