Skip to content
kyzhao edited this page Jul 14, 2014 · 5 revisions

Based on caffe im2col.Merge the loop into one kernel.

On GTX640:

  • image const int height = 256; const int width = 256; const int channels = 3; const int batch_size = 32;//128; const int ksize = 5; // 5-11 const int pad = 2; // 0-2 const int stride = 1; // 1 const int num_kernels = 64;

Clone this wiki locally