Open Access

Table A.1.

Characteristics of each layer of the CNN architecture.

Layer Inputs Kernel size h × w #feature maps
C1 Input image 5 × 5 64 × 64 64
P1 C1 2 × 2 (stride 2 pix) 32 × 32 64
C2, C3, C4 P1 1 × 1, 1 × 1, 1 × 1 32 × 32, 32 × 32, 32 × 32 48, 48, 48
C5 P1 1 × 1 32 × 32 64
C6 C2 3 × 3 32 × 32 64
C7 C3 5 × 5 32 × 32 64
P2 C4 2 × 2 (stride 1 pix) 32 × 32 48
Co1 32 × 32 240
C8, C9, C10 Co1 1 × 1, 1 × 1, 1 × 1 32 × 32, 32 × 32, 32 × 32 64, 64, 64
C11 Co1 1 × 1 32 × 32 92
C12 C8 3 × 3 32 × 32 92
C13 C9 5 × 5 32 × 32 92
P3 C10 2 × 2 (stride 1 pix) 32 × 32 64
Co2 32 × 32 340
P4 Co2 2 × 2 (stride 2 pix) 16 × 16 340
C14, C15, C16 P4 1 × 1, 1 × 1, 1 × 1 16 × 16, 16 × 16, 16 × 16 92, 92, 92
C17 P4 1 × 1 16 × 16 128
C18 C14 3 × 3 16 × 16 128
C19 C15 5 × 5 16 × 16 128
P5 C16 2 × 2 (stride 1 pix) 16 × 16 92
Co3 16 × 16 476
C20, C21, C22 Co3 1 × 1, 1 × 1, 1 × 1 16 × 16, 16 × 16, 16 × 16 92, 92, 92
C23 Co3 1 × 1 16 × 16 128
C24 C20 3 × 3 16 × 16 128
C25 C21 5 × 5 16 × 16 128
P6 C22 2 × 2 (stride 1 pix) 16 × 16 92
Co4 16 × 16 476
P7 Co4 2 × 2 (stride 2 pix) 8 × 8 476
C26, C27 P7 1 × 1, 1 × 1 8 × 8, 8 × 8 92, 92
C28 P7 1 × 1 8 × 8 128
C29 C26 3 × 3 8 × 8 128
P8 C27 2 × 2 (stride 1 pix) 8 × 8 92
Co5 8 × 8 348
FC1, FC2 Co5, FC1 1024, 1024

Notes. Columns are: name of the layer, input layer, size of the convolution kernel (in pixels), size (height × width in pixels) and number of the resulting feature maps.

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.