Cross-modal (text and figures) Analysis of a Scientific Corpus from Semantic Scholar - model weights

In the related notebook we show the application of cross-modal techniques to improve the categorization of scientific papers through content related to figures, using both the textual part (captions) and the visual part (figures, diagrams, images) jointly. To this purpose, we use several CNN models and execute some experiments, illustrating our approach. This deposit contains the model weights trained on such large corpora.