Table 3.
Runtime performance of TensorHolo V2 RGB-D CNNs on GPU and iPhone 13 Pro
| 30 layers | 15 layers | 8 layers | |
|---|---|---|---|
| NVIDIA TITAN RTX | 40 ms | 25 ms | 17 ms |
| iPhone 13 Pro | 478 ms | 297 ms | 209 ms |
The numbers are reported for the CGH synthesis networks with a different number of convolution layers. When the target hologram plane offsets from the midpoint hologram, the addition of the pre-filtering CNN adds 4.8 ms for a TITAN RTX and 60 ms for an iPhone 13 Pro.