Free Access

Table B.1.

Network architecture used in our experiments.

Layer # filters Function
Conv2D+ReLU 1 n Conv 3 × 3 + ReLU
Conv2D+ReLU 2 n Conv 3 × 3 + ReLU
MaxPooling 1 n Maxpooling 2 × 2
Conv2D+ReLU 3 n × 2 Conv 3 × 3 + ReLU
Conv2D+ReLU 4 n × 2 Conv 3 × 3 + ReLU
MaxPooling 2 n × 2 MaxPooling 2 × 2
Conv2D+ReLU 5 n × 4 Conv 3 × 3 + ReLU
Conv2D+ReLU 6 n × 4 Conv 3 × 3 + ReLU
Upsampling 1 n × 4 Upsampling 2 × 2
Conv2D+ReLU 7 n × 2 Conv 3 × 3 + ReLU
Concatenate 1 n × 2 Concat with Conv2D+ReLU 4
Conv2D+ReLU 8 n × 2 Conv 3 × 3 + ReLU
Conv2D+ReLU 9 n × 2 Conv 3 × 3 + ReLU
Upsampling 2 n × 2 Upsampling 2 × 2
Conv2D+ReLU 10 n Conv 3 × 3 + ReLU
Concatenate 1 n Concat with Conv2D+ReLU 2
Conv2D+ReLU 11 n Conv 3 × 3 + ReLU
Conv2D+ReLU 12 n Conv 3 × 3 + ReLU
Network Output m Conv 1 × 1

Notes. Number of network input feature maps n and the dimension of the output m depend on the experiment.

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.