Table 3 Comparison towards the floating point computations (FLOPs) when processing images with the resolution of \(1216\times 912\). Only the single-precision floating point multiplication are taken into account
From: TasselNetv2: in-field counting of wheat spikes with context-augmented local regression networks
| Â | TasselNet | TasselNetv2 | |
|---|---|---|---|
Non-overlap | Dense sample | ||
conv1 | \(4.70\times 10^{8}\) | \(6.92\times 10^{9}\) | \(4.79\times 10^{8}\) |
conv2 | \(1.24\times 10^{9}\) | \(1.83\times 10^{10}\) | \(1.28\times 10^{9}\) |
conv3 | \(1.22\times 10^{9}\) | \(1.81\times 10^{10}\) | \(1.28\times 10^{9}\) |
conv4 | \(2.44\times 10^{9}\) | \(3.61\times 10^{10}\) | \(2.56\times 10^{9}\) |
conv5 | \(2.44\times 10^{9}\) | \(3.61\times 10^{10}\) | \(2.56\times 10^{9}\) |
conv6(fc1) | \(5.17\times 10^{8}\) | \(2.07\times 10^{9}\) | \(2.07\times 10^{9}\) |
conv7(fc2) | \(1.75\times 10^{7}\) | \(6.46\times 10^{7}\) | \(6.46\times 10^{7}\) |
conv8(fc3) | \(1.26\times 10^{5}\) | \(5.05\times 10^{5}\) | \(5.05\times 10^{5}\) |
Total | \(8.34\times 10^{9}\) | \(1.16\times 10^{11}\) | \(1.03\times 10^{10}\) |