Last May, Richard Zhang, Philip Isola, and Alexei A. Efros from the University of Berkeley Vision Lab published their work “Colorful Image Colorization.” This paper describes a novel use of a convolutional neural net (learn more about deep learning) for colorizing black and white pictures.
We first heard of their work through the academic paper aggregator arxiv.org and then the accompanying GitHub repository where they made the model available.
We contacted the university research team about making their deep learning model available as a scalable microservice on Algorithmia, where more than 30,000 developers could then use and integrate their research into apps and services.
Algorithmia provides a common API for algorithms, functions, and models that run as scalable, serverless microservices. Colorful Image Colorization is one of 3,000 microservices available to the community of developers. Algorithmia is the largest algorithm marketplace in the world, which means the team could also earn money for their work based on the real-world application of their research. More on this below.
As part of a joint effort to promote their innovative work, we created Colorize-It.com, an interactive demo powered by the team’s research running as a microservice on Algorithmia.
More than 500,000 people had interacted with the microservice to colorize over 1M images in just a few days. On social media, over 50,000 people shared the demo.
Within a week, more than 200 different articles in 41 languages had been written about the Berkley Vision Lab’s research, representing a reach of over 10 million people.
On the marketplace side, in less than a week, the colorizer microservice had earned just under $1,000 dollars, which allowed for the purchase of a brand new Titan X GPU so they could continue doing their research.
Colorful Image Colorization is a great example of groundbreaking university research being made accessible and discoverable to developers across the globe through the Algorithmia platform. It made for easy integration into applications by third-party developers, and the researchers at the UC Berkeley Vision Lab never had to worry about configuring infrastructure or maintaining the service as usage scaled up.
We continue to work Zhang, Isola and Efros and are excited to bring more of their research to developers everywhere.