Fig. 2.

Training procedure of the progressively growing architecture (CHRONNOS). The next resolution level is gradually faded in by separating the input and output layers (yellow) of the original core model into a separate branch (old branch). The new branch introduces new ConvBlocks (blue) and is faded in by increasing the α parameter over the training cycle. The difference in resolution is adjusted by up- and down-sampling layers for the old branch and by convolutional layers for the new branch (green). Once the old branch is faded out (α = 1), the old branch is removed and the remaining architecture is stabilized by an additional training cycle. The new branch and core model form the core model for the next resolution level.
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.