Automating crowdsourced video creation with Google Cloud Platform and Machine Learning APIs

By making it easy for companies to create crowd-sourced video, award-winning London startup Seenit has won major clients including Rolls-Royce, Unilever and the BBC in just three years. Created in 2014, Seenit uses its app and online studio to provide a platform for crowd-sourced footage. Sophisticated APIs automatically assess and tag each clip as it’s uploaded, creating detailed metadata that producers can search to find the video they need. That makes accuracy and processing speed vital to the company’s offering. Seenit found both with Google Cloud Platform and Google Cloud Machine Learning Platform.

“The quality of the results are ultimately what define our success, and the quality of data we get from Google APIs mean that search results are immediately very relevant to what people need. I ran benchmarks on AWS, Azure and GCP. For our use case, GCP’s network and disk speeds were up to twice as fast as the competition.” - Dave Starling, CTO, Seenit

Sorting content – saving time

Film editors typically review every clip they receive before assembling a finished video. This creates long and costly turnarounds for crowd-sourced footage, which may consist of thousands of videos that take dozens of hours to watch and sort. Seenit aims to make crowdsourcing video quick and affordable by automating this review process, so films can be turned around at great speed and low cost.

To do this, Seenit chose Google Cloud Platform. Google Cloud Machine Learning APIs including Vision, Speech, and Natural Language help sort clips by objects in shot, gender of speakers, sentiment and speech as they’re uploaded. Seenit receives new APIs from Google in alpha or beta stages as a trusted tester, and because they fit straight into GCP, they can be up and running in less than a day, keeping the company at the cutting edge of developments. The company maintains optimal performance by managing incoming videos on RabbitMQ and using Compute Engine to autoscale between two and 20 servers. And because everything is on one platform, processes are not slowed down by calling third parties.


“The great thing with Google Cloud Machine Learning Platform is that it gets better over time, as the more people use it, the bigger the training set. Originally we set our confidence threshold for results from the APIs at around 80 percent, because we wanted to be really sure. After three months we realised we could drop it to 60 percent. It noticeably improves over time.” - Dave Starling, CTO, Seenit

Automating editing – scaling worldwide

Seenit won the Startup Battlefield at TechCrunch Disrupt London 2016 for its successful innovation and outstanding client list. This year the company is looking to propel growth by automating and scaling more features, such as algorithmic film editing and daily video digests made from crowdsourced footage. It is also looking to expand its services to the US and Asia, and because everything is on GCP, the entire Seenit platform can be duplicated in any location within a few minutes on 12 VMs. “From committing code,” says Starling, “we can have it live in minutes and scaled in seconds.”

“The product teams at Google are great, because we can feed stuff back to them and they're not precious about it. It helps them improve the product. With AWS, you'd have to be spending a lot more money to get that level of input. The GCP team is more interested in making a better product than how much you're paying.” - Dave Starling, CTO, Seenit