1

How do I relate compute complexity in Convolutional Neural Network to processing time of one image in object detection for a given CPU/GPU's processing power?

Say my CNN architecture needs 'parameters to tuned': 20984, 'compute_complexity': 0.039456 GFLOPs, 'memory_req': 12.109375MB.

Then I have a workstation with processor Intel® Xeon(R) CPU E5-1660 0 @ 3.30GHz × 12 and RAM 16G. Nvidia GPU is P4000 Quadro with 1792 GPU cores.

If i use CPU only in detection, how can I ESTIMATE processing time for one image in object detection.

If I use GPU, how can I ESTIMATE processing time for one image in object detection.

batuman
  • 441
  • 1
  • 3
  • 10
  • test it empirically? also, computational complexity is the generally study of how much time a computational procedure takes as a function of the size of the inputs, whereas CNNs are typically operated on a constant input size. – shimao Aug 17 '18 at 04:32
  • Thanks for the response. So computational complexity is varied from one network architecture to another. What would be the parameter to make assessment of a network's computation time for a given CPU/GPU resources – batuman Aug 20 '18 at 06:05
  • @shimao My CNN network has `compute_complexity: 0.039456 GFLOPs`. Then my `Intel® Xeon(R) CPU E5-1660 has compute capacity:4 GFLOPS`. For detection, single image is input. Can I estimate maximum one image processing time is `(0.039/4)secs`? – batuman Aug 20 '18 at 07:03

0 Answers0