Free Access
Table B.2.
Network architecture to calculate the variance.
| Layer | # filters | Function |
|---|---|---|
| Conv2D+ReLU 1 | n | Conv 3 × 3 + ReLU |
| Conv2D+ReLU 2 | n | Conv 3 × 3 + ReLU |
| Conv2D+ReLU 3 | n | Conv 3 × 3 + ReLU |
| Conv2D+ReLU 4 | n | Conv 3 × 3 + ReLU |
| Conv2D+ReLU 5 | n | Conv 3 × 3 + ReLU |
| Conv2D+ReLU 6 | n | Conv 3 × 3 + ReLU |
| Conv2D+ReLU 7 | n | Conv 3 × 3 + ReLU |
| Conv2D+ReLU 8 | n | Conv 3 × 3 + ReLU |
| Conv2D 13 | 1 | Conv 1 × 1 |
| Variance Output | – | −Abs(x)−12 |
Notes. Given that the variance in log units is a big negative number, to accelerate the convergence of the network we had to use an additional layer to constrain its value.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.