Augmenting text documents with images
The objective of this project is to provide illustration images for a text document (a text article for instance) automatically.
The project has two steps : First we find the topic of the text. Then we find images relevant in the image-net database.
- Improve the tf-idf : word frequency
- Improve the tf-idf : plural
- Be able to have a feed back on the images : select the good synsets and then fetch images in the hyponyms
- Clustering on colors: Resnet: res convu => output vect taille 2028 => mettre vec dans ACP/PCA (ou TSNE)
- Présenter résultats dans un notebook jupyter
- Requirements
- VirtualEnv
- Default values
(transfer learning => reduction dimmension => clusterisation => systeme de recommendation)