Free Access

Table 2.

LASTRO EPFL architecture.

Layer type Shape Activation # parameters
Convolutional 4 × 4 101 × 101 × 1/4 → 98 × 98 × 16 Rectifier 256/1024 + 16
Convolutional 3 × 3 98 × 98 × 16 → 96 × 96 × 16 Rectifier 2304 + 16
Max pool/2 96 × 96 × 16 → 48 × 48 × 16
Batch normalization 48 × 48 × 16 16 + 16
Convolutional 3 × 3 48 × 48 × 16 → 46 × 46 × 32 Rectifier 4608 + 32
Convolutional 3 × 3 46 × 46 × 32 → 44 × 44 × 32 Rectifier 9216 + 32
Max pool/2 44 × 44 × 32 → 22 × 22 × 32
Batch normalization 22 × 22 × 32 32 + 32
Convolutional 3 × 3 22 × 22 × 32 → 20 × 20 × 64 Rectifier 18 432 + 64
Convolutional 3 × 3 20 × 20 × 64 → 18 × 18 × 64 Rectifier 36864 + 64
Max pool/2 18 × 18 × 64 → 9 × 9 × 64
Batch normalization 9 × 9 × 64 64 + 64
Dropout 9 × 9 × 64
Convolutional 3 × 3 9 × 9 × 64 → 7 × 7 × 128 Rectifier 73 728 + 128
Dropout 7 × 7 × 128
Convolutional 3 × 3 7 × 7 × 128 → 5 × 5 × 128 Rectifier 147 456 + 128
Batch normalization 5 × 5 × 128 128 + 128
Dropout 5 × 5 × 128
Fully-connected 5 × 5 × 128 → 1024 Rectifier 3 276 800 + 1024
Dropout 1024
Fully-connected 1024 → 1024 Rectifier 1 048 576 + 1024
Dropout 1024
Fully-connected 1024 → 1024 Rectifier 1 048 576 +1024
Batch normalization 1024 1024 + 1024
Fully-connected 1024 → 1 Sigmoid 1024 + 1
Total ≈5 674 000

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.