Fig. 13
Download original image
Learned gamma (γ) weights for the weighted-average embedding. Each γi scales the contribution of the output from block i of the encoder (γ0 for the input). The classifier learns to assign higher importance to intermediate layers (2−5) rather than the initial or final layers, especially as more training data becomes available.
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.