The live viewer

It runs locally, not here

The viewer is an interactive application, not a page. It decodes film, foveates each frame at its own current fixation, runs a saliency model and a vision-language model on that percept, and samples its next fixation. That needs a Python server, a GPU for the two models, and roughly 700 MB of video and mesh assets, so it is not something a static site can host and not something that belongs in a repository.

To run it from the full project:

python paper-hub/server.py
# then open http://127.0.0.1:8077/viewer

Everything the viewer produced is on this site already: the figures are generated from its stored outputs, and the three documents report what it did. Back to the project.