SEEING SEEING: Reimagining MoMA’s Digital Collection as an Archive of Attention
2025
Harvard GSD |
ADV-9705| Reimagining the Archive
Developed in collaboration with the Museum of Modern Art, NYC
Museums often rely on textual metadata for cataloguing practices: artist, date, medium, and provenance. By tracking and analyzing viewers’ visual paths across digital images of artwork, this project proposes a new layer of embodied metadata, one that is subjective and interpretive rather than curatorially imposed. The digital archive may then be reconstructed through its own history of being viewed. What if the way we view archives could leave an imprint on the archive itself?
Computation, Visualization, Perception Studies, Art
This project focuses on several of MoMA’s “unmissable” artworks, according to the museum’s website. The physical artworks are of vastly different sizes and styles,
however the experience of viewing these images digitally reduces them all to similar scale, color, and overall visual quality.
One: Number 31, 1950 (1950)
Jackson Pollock
Oil & enamel paint on canvas
8’ 10”x17’ 5 5/8”
The Starry Night (1889)
Vincent van Gogh
Oil on canvas
29” x 36 1/4”
Le Demoiselles d’Avignon (1907)
Pablo Picasso
Oil on canvas
8’ x 7’ 8”
American People Series #20: Die (1967)
Faith Ringgold
Oil on canvas, two panels
72” x 144”
Water Lillies (1914-1926)
Claude Monet
Oil on canvas, three panels
Each 6’ 6 3/4” x 12’ 11 1/4”
To examine this digital viewing experience, I built a custom gaze tracking web interface using WebGazer.JS. After a short callibration exercise, viewers visual paths are tracked and recorded through their webcam over a 30 second viewing session.
The visual path is recorded as a line drawing, where lineweight correlates to duration of gaze lingering. The image is then divided into grid cells and a spatial heatmap is generated, where warmer colors indicate longer viewing times. Lastly, the image is regenerated with only the grid cells that were “hit” visible.
Gaze pathHeat mapViewed grid cells
All results can be viewed, toggled, and overlayed within the application. Statistics are generated for the percentage of image surface area viewed, grid cells with longest and shortest view times, and an average dwell time per image grid celll.
A demonstration of the interface after viewing Ringgold’s
American People Series #20: Die
Resutls were then compared across multiple viewing sessions, of the same image, to see how visual attention persists or shifts over time.
1 2 3
Viewing sessions can also be layered to “develop” the image over time, with the most opaque areas being the most viewed.
Session 1Session 1 & 2 compositeSession 1, 2, & 3 final composite
Layering can be also done with RGB additive color mixing. Only the areas that were viewed in all 3 sessions appear in full color in the composite image.
Session 1 masked in blueSession 1 & 2 (green) compositeSession 1, 2, & 3 (red) final RGB composite
Results from Monet’s Water Lillies
Results from Pollock’s
One: Number 31, 1950
Results from Ringgold’s
American People Series #20: Die
A sampling of other results, images, and outtakes from the process
PREVIOUS