A Slice of MoMA
is an interactive data visualization of the Museum of Modern Art’s online collection. It proposes a playful way of visualizing a large amount of artwork data by approaching them from a color analysis perspective. It stimulates visitors to explore and compare artists, reveals the color progression of their artworks, and recommends artworks of similar visual styles based on visitors interests.
The Museum of Modern Art in NYC houses 200,000 pieces of modern and contemporary art works. Among them, 81,000 works are accessible to the public digitally through a search engine on MoMA’s website.
MoMA's Current Website: https://www.moma.org/collection/
MoMA's Current Website: https://www.moma.org/collection/
We started the project by looking at MoMA’s current webpage of displaying its digital collection, and identified the following problems:
It is mainly designed for professionals who know exactly what they want to search for
It does not give a good overview of MoMA’s digital collection
According to a survey conducted by MoMA, visitors of the website report medium to high art knowledge and are likely to be professionally engaged with the subjects. The current website clearly doesn’t fulfill the goal of reaching a wider audience group and promoting discovery.
Design Goals
As a result, we have defined the following design goals for our project:
Make MoMA’s online collection more accessible to people who don’t have much background knowledge about modern art
Transform the traditional way of displaying artworks online to an inspirational exploration tour
Method & Process
Design Strategies
We approached our design problem with the following strategies:
1. Focus on extracting artworks’ color
Color is something easy to understand and people can associate themselves with. It is also something that grabs people’s attention when during first encountering. Therefore, we’ve explored different methods to extract the color information from the artwork, and use that to direct the whole experience.
2. Progressive revealing layers of information
We were dealing with a large dataset that have 80,000+ entries of artworks and 10000+ entries of artists. In order to show both high-level overview as well as details of each piece, we’ve looked into ways to break down the exploration into a step-by-step process, with each step revealing a bit more information.
3. Exploratory path through association
Last but not least, we want to avoid the traditional way of display artworks in a grid layout, but allow user to have a flexible path of navigation based on their interests. We wish user would be able to explore artworks through association instead of indexing, which led us to build an artwork recommendation system powered by machine learning.
References
We studied a series of inspiring data visualizations for reference.
Bohemian Bookshelf
The Bohemian Bookshelf is a bookshelf visualization that was designed to facilitate open-ended exploration of digital library collections and serendipitous book discoveries.
On Broadway
On Broadway folds Google street view maps and Instagram posts of street segments as slices. Users can click to expand the whole images and see more details.
Front Row to Fashion Week
In this visualization created for 2013 New York Fashion Week, dazzling collections were presented in carousels (one for each designer). All images are really thin, only showing a glimpse of the main colors of the outfit until user mouseover it.
Iterative Process
We went through visual encoding and design iterations.click to read more
We started by looking at the type of information that are available in the dataset, what we want to present, and the best visual encoding.
Meanwhile, we also started to making hand sketches as well as mockups in Illustrator.
On the technical side, we researched 3 different ways to build the artwork recommendation systemclick to read more
There are a few directions we thought about to compute similarities between artworks:
based on artwork’s metadata like artist, year of creation, medium, genre
based on visual features extracted from photo of artworks (it works because most of the artwork in our dataset are 2d like paintings or drawings)
based on artwork’s description
We landed on the first option. More specifically, we used a pre-trained VGG19 model to compute a score of similarity between each pair of artworks, and build the art recommendation system based on that score. (Please check out our
Github repo
if you are interested in the technical details)
We also iterated on the color analysis algorithm to make our graph more appearingclick to read more
As the first piece of the data visualization, we’ve designed a bar graph to showcase the representative colors of all artists. However, we didn’t realize until the implementation that the graph is not as colorful as we have expected. It is because of the way we calculate those color: we wanted to use the dominant colors which have the highest area ratio on the canvas in artworks. However, the colors turned out to be dark and grayish. The saturation for all the colors is remarkably low. It is partially due to the fact that modern art has a prevailing preference for low-saturated colors, but also that the thumbnail images we used for color analysis sometimes include large areas of margins. The resulting graph has a quite limited visual expressiveness.
Bar graph that showcases each artist's representative color before we change the color analysis algorithm
To solve this , in the end we have chosen to use the colors with the highest saturation from clustering the major colors in each artworks. The refined color palette is much easier to read and distinguish, and adds a lot more vitality to the visualization.
Bar graph that showcases each artist's representative color after we change the color analysis algorithm
Deliverable
Please feel free to play with our live prototype, which we coded from scratch with HTML/CSS, JavaScript, using D3.js and Boothstrap. Please note the prototype only works for desktop.
If you are on mobile, please feel free to check out this demo video
Retrospect
A Slice of MoMA is our first attempt to explore the concept of online museum and displaying artwork datasets in a more exploratory and interactive way. Working on it as the final project of a data-visualization class, we had to prioritize tasks within the limited timeframe. If we had more time and resources to work on this project, I wish we had:
send out survey in the beginning to better understand what people want when they browse art museum’s digital collection, and run usability tests after we build our prototype
take responsiveness and accessibility into considerations
utilize existing JavaScript UI frameworks to build the prototype more efficiently