Practical Deep Learning for Cloud, Mobile, and Edge

Chapter 8. Cloud APIs for Computer Vision: Up and Running in 15 Minutes

Due to repeated incidents of near meltdown at the nearby nuclear power plant, the library of the city of Springfield (we are not allowed to mention the state) decided that it was too risky to store all their valuable archives in physical form. After hearing that the library from their rival city of Shelbyville started digitizing their records, they wanted to get in on the game as well. After all, their collection of articles such as “Old man yells at cloud” and “Local man thinks wrestling is real” and the hundred-year-old iconic photographs of the Gorge and the statue of the city’s founder Jebediah Springfield are irreplaceable. In addition to making their archives resilient to catastrophes, they would make their archives easily searchable and retrievable. And, of course, the residents of Springfield now would be able to access all of this material from the comfort of their living room couches.

The first step in digitizing documents is, of course, scanning. That’s the easy part. Then starts the real challenge—processing and understanding all of this visual imagery. The team in Springfield had a few different options in front of them.

• Perform manual data entry for every single page and every single photograph. Given that the city has more than 200 years of rich history, it would take a really long time, and would be error prone and expensive. It would be quite an ordeal to transcribe all of that material. ...