You can obtain the model FLOPs count and the amount of memory consumed using the model.getFLOPs and model.getMemoryConsumption methods. Both methods take as input the specified blob shape. Per-layer inference time statistics are available after the forward pass is performed and can be obtained via the model.getPerfProfile method, which returns total inference time and per-layer timings, all in ticks.
The following output is expected:
gflops: 3.1904431360000003weights (mb): 27.994208 , blobs (mb): 45.92096inference (ms): 83.478832LAYER TIME (ms)conv1/7x7_s2 4.57conv1/relu_7x7 0.00pool1/3x3_s2 0.74pool1/norm1 1.49conv2/3x3_reduce 0.57conv2/relu_3x3_reduce 0.00conv2/3x3 11.53conv2/relu_3x3 0.00conv2/norm2 3.35pool2/3x3_s2 0.90 ...
