Open Access

Table 4

Details of encoder and decoder architecture for 96 × 96 pixel inputs.

Encoder layers Decoder layers


Layer In Ch. Out Ch. Kernel Stride Padding Layer In Ch. Out Ch. Kernel Stride Padding
PCDown1 1 32 3 1 1 PCUp1 512 512 3 1 0
PCDown2 32 64 4 2 1 PCUp2 1024 512 4 2 1
PCDown3 64 64 3 1 1 PCUp3 1024 512 3 1 1
PCDown4 64 128 4 2 1 PCUp4 1024 256 4 2 1
PCDown5 128 128 3 1 1 PCUp5 512 256 3 1 1
PCDown6 128 256 4 2 1 PCUp6 512 128 4 2 1
PCDown7 256 256 3 1 1 PCUp7 256 128 3 1 1
PCDown8 256 512 4 2 1 PCUp8 256 64 4 2 1
PCDown9 512 512 3 1 1 PCUp9 128 64 3 1 1
PCDown10 512 512 4 2 1 PCUp10 128 32 4 2 1
PCDown11 512 512 3 1 0 PCUp11 64 1 3 1 1

Notes. The encoder consists of 11 PCDown blocks, each employing a ReLU activation function. The decoder is composed of 11 PCUp blocks, using leaky ReLU activation function for all but the final block.

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.