Fig. 6

Download original image
Overall model framework, divided into three main parts. The input part is a high-dimensional matrix with dimensions of 48 × 48 × 5. The feature extraction consists of five stages, with the Swin Transformer block (in blue) detailed in Sect. 3.2.2, and the patch merging (in green) described in Sect. 3.2.1. Numbers within the Swin Transformer blocks denote the repetition of stacked blocks. The third part is the head for the classification output, where the Bayesian block (see Fig. 7) maps the 8C-dimensional features to 2 or 3, corresponding to two-class and three-class classification, while also outputting the uncertainty.
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.