Model params 194 MB
Estimates for a single full pass of model at input size 600 x 850:
- Memory required for features: 2 GB
- Flops: 117 GFLOPs
Estimates are given below of the burden of computing the res5c_relu
features in the network for different input sizes using a batch size of 128:
input size | feature size | feature memory | flops |
---|---|---|---|
300 x 425 | 19 x 27 x 2048 | 55 GB | 3 TFLOPs |
600 x 850 | 38 x 53 x 2048 | 218 GB | 13 TFLOPs |
900 x 1275 | 57 x 80 x 2048 | 493 GB | 29 TFLOPs |
1200 x 1700 | 75 x 107 x 2048 | 871 GB | 51 TFLOPs |
1500 x 2125 | 94 x 133 x 2048 | 1 TB | 80 TFLOPs |
1800 x 2550 | 113 x 160 x 2048 | 2 TB | 116 TFLOPs |
A rough outline of where in the network memory is allocated to parameters and features and where the greatest computational cost lies is shown below. The x-axis does not show labels (it becomes hard to read for networks containing hundreds of layers) - it should be interpreted as depicting increasing depth from left to right. The goal is simply to give some idea of the overall profile of the model: