Table A.1
Characteristics of each layer of the CNN architecture.
Layers | Inputs | Kernel size | h × w | #Feature maps |
---|---|---|---|---|
P1, P2 | LCI | 5 × 1, 11 × 1 (stride 2) | 850 × 5 | 1, 1 |
TC1, TC2, TC3 | P1 | 11 × 1, 21 × 1, 41 × 1 | 850 × 5 | 16 16, 16 |
TC4, TC5, TC6 | P2 | 11 × 1, 21 × 1, 41 × 1 | 850 × 5 | 16, 16, 16 |
MC1 | C1 | 1 × 5 | 850 × 1 | 96 |
TC7, TC8 | MC1 and C1 | 11 × 1, 21 × 1 | 850 × 5 or 850 × 1 | 24, 24 |
P3 | C2 | 3 × 1 (stride 2) | 425 × 1 | 48 |
P4 | C3 | 3 × 1 (stride 2) | 425 × 5 | 48 |
P5 | LCI | 21 × 1 (stride 4) | 425 × 5 | 1 |
TC9, TC10, TC11 | P5 | 5 × 1, 11 × 1, 21 × 1 | 425 × 5 | 16, 16, 16 |
TC12, TC13 | C4 | 11 × 1, 21 × 1 | 425 × 5 | 24, 24 |
MC2 | C5 | 1 × 5 | 425 × 1 | 48 |
TC14 | C6 and C7 | 11 × 1 | 425 × 1 or 425 × 5 | 96, 96 |
P6 | TC14 | 3 × 1 (stride 2) | 212 × 1 | 96 |
P7 | TC14 | 3 × 1 (stride 2) | 212 × 5 | 96 |
TC15 | P6 and P7 | 11 × 1 | 425 × 1 or 425 × 5 | 48, 48 |
P8 | TC15 | 3 × 1 (stride 2) | 106 × 1 | 48 |
P9 | TC15 | 3 × 1 (stride 2) | 106 × 5 | 48 |
P10 | LCI | 41 × 1 (stride 16) | 106 × 5 | 1 |
TC16, TC17, TC18 | P10 | 5 × 1, 11 × 1, 21 × 1 | 106 × 5 | 16, 16, 16 |
TC19 | C8 | 11 × 1 | 106 × 5 | 48 |
MC3 | TC19 | 1 × 5 | 106 × 1 | 48 |
TC20 | C9 and C10 | 11 × 1 | 106 × 1 or 106 × 5 | 128, 128 |
P11 | TC20 | 3 × 1 (stride 2) | 53 × 1 | 48 |
P12 | TC20 | 3 × 1 (stride 2) | 53 × 5 | 48 |
P13 | LCI | 61 × 1 (stride 32) | 53 × 5 | 1 |
TC21, TC22, TC23 | P13 | 5 × 1, 11 × 1, 21 × 1 | 53 × 5 | 16, 16, 16 |
TC24 | C11 | 11 × 1 | 53 × 5 | 64 |
TC25, TC26 | C12 | 11 × 1, 21 × 1 | 53 × 5 | 64, 64 |
MC4 | C13 | 1 × 5 | 53 × 1 | 64 |
TC27, TC28, TC29, TC30 | c rowspan="0" c>0 * [0pt]C14 | c rowspan="0" c>0 * [0pt]5 × 1, 11 × 1, 21 × 1, 41 × 1 | c rowspan="0" c>0 * [0pt]53 × 1 | c rowspan="0" c>0 * [0pt]48, 48, 48, 48 |
FC1, FC2 | C15, FC1 | – | – | 1024, 1024 |
Notes. The table lists: the name of the layer, the input layer, the size of the convolution kernel (in pixels), the size in pixels (height × width) of resulting feature maps and the number of resulting feature maps. The concatenation layers are not represented here but they are present in Fig. 4.
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.