| | 12 | </p> |
| | 13 | |
| | 14 | <p> |
| | 15 | 189. |
| | 16 | <b>Make Gaussian filter more memory efficient.</b><br> |
| | 17 | Mikyung Han<br> |
| | 18 | June 17, 2009<br> |
| | 19 | Make Gaussian filter more memory efficient. Currently it uses numpy fft library |
| | 20 | which only provides double precision, and can't transform in place. So 16 bytes |
| | 21 | per voxel are required. For single byte data that is 16x bigger than the input |
| | 22 | data. Could transform a plane at a time building 32-bit float array. This would |
| | 23 | also permit progress messages and canceling. Other improvements would be distributing |
| | 24 | 64-bit Mac Chimera or using fftw for fourier transforms. |