Table 2.
Reconstruction times in fps for various image sizes with side length N and fixed versus variable oversampling ratio γ. For some input sizes, 1.5 is the optimal value (N = 128, N = 144); for other sizes, the number of grid sample locations must be increased to yield a speed-up (N = 160, N = 170). For a third group of image sizes (N = 256), the optimal FFT size increases the size of the data to a degree that the speed-up is reduced or even nullified by additional overhead. For each test case, 200 frames were reconstructed using 1 accelerator.
N | Fixed γ = 1.5 | Optimal γ ≥ 1.4 | S | |||
---|---|---|---|---|---|---|
G | fps | γ | G | fps | ||
128 | 384 | 8.1 | 1.5 | 384 | 8.3 | 1.02 |
144 | 432 | 2.5 | 1.5 | 432 | 2.5 | 1.00 |
160 | 480 | 4.4 | 1.51875 | 486 | 5.0 | 1.25 |
170 | 510 | 2.5 | 1.50588 | 512 | 5.8 | 2.32 |
256 | 768 | 2.4 | 1.53125 | 784 | 2.4 | 1.00 |