DV DataReel: The data behind every frame

DV DataReel

This site shows what can be measured automatically in a collection of 2,735 films. Click around and explore for yourself.

What was measured

Every film was cut into shots. Each shot was then measured for its length, how close the camera stands to the people in it, how many people are on screen, how much the camera moves, and how much of it is speech or music.

A language model also read each film from its dialogue, its title cards or a summary of its plot, divided it into scenes and described each one. Every scene has a contact sheet of frames taken from its shots.

The faces in each film were tracked from shot to shot and matched to the credited cast, by a language model comparing them with the cast's photographs and by their likeness to those photographs.

The corpus page

The posters and the plot show the same films. Narrow them with the filters above, change the axes and colour of the plot, and sort the films by any measure. Switch to the list to see each film's value beside its title. Hover a poster or a row to find its dot.

Searching the scenes

The pink box searches every scene description by what it means and by the words in it, and ranks the films by their best scene. Each film in the list quotes that scene beside its contact sheet. On a film page the scenes are shaded by how well they match. Put a phrase in quotes to require it as written. The search runs entirely in your browser, and its model and index load on the first search.

A film page

Click a film to see its scenes with their descriptions and contact sheets, who is on screen and when, its shots laid out over its runtime, and where it stands against the rest of the corpus.

films sort by?

color

characters onscreen ?