For the final project, I’m working with a set of 3300 poster images from the Cooper Hewitt collection. Some useful metadata is provided for these posters such as year of origin, year of acquisition, title, description and color palette. A few of my original ideas are mentioned below (crossed-out if entirely eliminated):

1. t-SNE by decades
2. object recognition
3. generate captioning
4. visualise one particular type of posters (advertisements, propaganda, movies)
5. contrast two different categories
6. generate posters by decade or topic

Currently, I’m moving in a somewhat different direction which is to create an interactive visualization (possibly a website) for all these posters using the available metadata. This could have basic ways of presenting the data such as sorting, filtering, searching etc. (using some of the initial ideas mentioned above) or more interesting ways that I plan to explore in the next few days.

Links:

• Cooper Hewitt poster collection