Table A.1.
Details of the architecture the CNN used to predict maps of treion(r).
| Network branch | Layer number | Layer type | Nbr of filters/data | Size of filter/data | Activation function |
|---|---|---|---|---|---|
| Encoder | 1 | Input | 31 500 | 128x128 | . |
| 2 | Conv2D+Max Pooling | 32 | 3x3 | Relu | |
| 3 | Conv2D+Max Pooling | 64 | 3x3 | Relu | |
| 4 | Conv2D+Max Pooling | 128 | 3x3 | Relu | |
| 5 | Conv2D+Max Pooling+Dropout | 256 | 3x3 | Relu | |
| 6 | Conv2D+Dropout | 512 | 3x3 | Relu | |
| Decoder | 7 | UpSampling+Merge+Conv2D | 256 | 3x3 | Relu |
| 8 | UpSampling+Merge+Conv2D | 128 | 3x3 | Relu | |
| 9 | UpSampling+Merge+Conv2D | 64 | 3x3 | Relu | |
| 10 | UpSampling+Merge+Conv2D | 32 | 3x3 | Linear | |
| 11 | Output | 31 500 | 128x128 | . | |
Notes. Each convolution layer within the encoder part is followed by a Max Pooling layer, except the sixth. Instead, each convolution layer within the decoder part is followed by an up-sampling layer plus a Merge layer that concatenates layers of same dimension of the encoder part with the corresponding layer of the decoder.
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.