Filter by tag

Introducing Machine Learning Assisted Video Object Tracking

In just three years, a million minutes of video content will cross global IP networks every second. That’s over 100 billion minutes a day. And if that sounds staggering, well, that’s because it is.

The thing is, according to Gartner, 99% of this video content is going to be analyzed by machines, not humans. Understanding this video, however, requires smart models and smart models require high-quality training data. That’s why we’re thrilled to announce the newest solution on Figure Eight’s human-in-the-loop platform: Machine Learning assisted Video Object Tracking.

Here’s how it works, appropriately enough, in video form:

Our new video object tracking solution combines human intelligence and machine learning to annotate video up to 100 times faster than human-only approaches. How exactly is the solution this fast? Let us explain:

You upload a video and create an ontology of what objects you’d like annotated. You can have up to 255 classes, each with dozens of labels per frame. Our system breaks the video down frame-by-frame and you launch your Figure Eight job.

Annotators will see the first frame of the video and will place bounding boxes over the objects you’ve requested. This process for first frame, functionally, behaves like an image annotation job. What happens next is an ensemble deep learning model kicks in. The labels from the first frame persist to the second, with that deep learning model tracking where they’ve moved in the next frame. Instead of re-annotating, contributors are there to simply spot-check, adjust, or resize the boxes now that those objects have moved in time and space. The model persists the labels so they don’t need to do that work again. Our solution also offers the ability to mark objects as occluded if they happen to move behind something or move out of frame for a few seconds as well.

This process continues each and every frame until the video is complete, making video annotation up to 100 times faster than human-only approaches.

We’re really proud of this new Figure Eight capability and are excited to release it to the world. If you’re looking to do video annotation, for any use case or industry, we’d love to show it off. You can check out our datasheet or just get in touch and we’ll walk you through everything.

Justin Tenuto

Justin Tenuto

Justin writes things at Figure Eight. He enjoys books about robots and aliens, baking bread, and is pretty sure he can beat you at pingpong.