Table B.1.
VGGNET-like model configuration.
Layer | Output shape | Params # |
---|---|---|
Input layer | (64, 64, NC) | 0 |
Conv2D | (64, 64, 64) | 6976 |
Leaky ReLU | (64, 64, 64) | 0 |
Conv2D | (64, 64, 64) | 36928 |
Leaky ReLU | (64, 64, 64) | 0 |
Max Pool2D | (32, 32, 64) | 0 |
Conv2D | (32, 32, 128) | 73856 |
Leaky ReLU | (32, 32, 128) | 0 |
Conv2D | (32, 32, 128) | 147584 |
Leaky ReLU | (32, 32, 128) | 0 |
Max Pool2D | (16, 16, 128) | 0 |
Conv2D | (16, 16, 256) | 295168 |
Leaky ReLU | (16, 32, 256) | 0 |
Conv2D | (16, 16, 256) | 590080 |
Leaky ReLU | (16, 16, 256) | 0 |
Conv2D | (16, 16, 256) | 590080 |
Leaky ReLU | (16, 32, 256) | 0 |
Conv2D | (16, 16, 256) | 590080 |
Leaky ReLU | (16, 16, 256) | 0 |
Max Pool2D | (8, 8, 256) | 0 |
Conv2D | (8, 8, 512) | 1180160 |
Leaky ReLU | (8, 8, 512) | 0 |
Conv2D | (8, 8, 512) | 2359808 |
Leaky ReLU | (8, 8, 512) | 0 |
Conv2D | (8, 8, 512) | 2359808 |
Leaky ReLU | (8, 8, 512) | 0 |
Conv2D | (8, 8, 512) | 2359808 |
Leaky ReLU | (8, 8, 512) | 0 |
Max Pool2D | (4, 4, 512) | 0 |
Conv2D | (4, 4, 512) | 2359808 |
Leaky ReLU | (4, 4, 512) | 0 |
Conv2D | (8, 8, 512) | 2359808 |
Leaky ReLU | (4, 4, 512) | 0 |
Conv2D | (4, 4, 512) | 2359808 |
Leaky ReLU | (4, 4, 512) | 0 |
Conv2D | (4, 4, 512) | 2359808 |
Leaky ReLU | (4, 4, 512) | 0 |
Max Pool2D | (2, 2, 512) | 0 |
Flatten | (2048) | 0 |
Dense | (4096) | 8392704 |
Leaky ReLU | (4096) | 0 |
Dropout | (4096) | 0 |
Dense | (4096) | 16781312 |
Leaky ReLU | (4096) | 0 |
Dropout | (4096) | 0 |
Dense | (2) | 8194 |
Output Layer | (2) | 0 |
Notes. The columns specify the layer operation, the shape of the output and the number of parameters to fit. The output shape of a layer is a 4-D matrix, but, since the first dimension is the fixed size of the input data batch (with a size of 64 patterns), we do not mention this number to prevent confusion. The total amount of trainable parameters is larger than 45M. The last dimension of the input layer is the involved number of channels (i.e. the number of photometric bands used), a quantity depending on the specific experiment (see Sect. 2).
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.