02.6.19

Introducing Objectbox — Super Simple Object Detection

We are extremely excited to announce the release of a brand new box from Machine Box. Introducing Objectbox, the super-simple, fully trainable, object detection solution…in a box.

Now, you can tackle use cases like;

  • Tracking the prevalence and value of brands and logos across all your media content.
  • Counting the number of people walking by your booth at a trade show.
  • Improving OCR results by selecting specific sections of your video to process.

Objectbox is 100% trainable.

You start by drawing some boxes around the objects you want to detect in images or video. Objectbox will give you feedback immediately so you know if your detector is improving or not.

Objectbox will be trained exclusively on your data so it’ll have the best chance of being an accurate object detector.

Getting started with Objectbox couldn’t be easier:

  1. Visit https://machinebox.io and click sign up.
  2. Follow the instructions to login via e-mail.
  3. Scroll down on the account page and copy and paste the Terminal command for Objectbox.

The only special thing you need to do that is different than Machine Box’s other boxes is designate a folder on your hard drive for the videos and images you want to use for training.

Drop some videos into the folder, spin up Objectbox, and click on “annotation tool”. You’ll then be able to draw boxes around things and give them labels.

After you draw the box, you can click train. Give it a few seconds to train the detector, then move the play head to another part in the video and watch as it attempts to detect the object.

Logo detected after one example

With this immediate feedback, you can quickly add more samples to improve results, or test other videos to see how well your detector works in other contexts.

Like our other boxes, once you’ve trained the model, it is easy to scale and deploy into production since everything is already a Docker container. Simply save your state file to scale the learning to other boxes.

If you’re an aiWARE user, you can also use Objectbox. Train a detector locally on your machine, upload the state file to an S3 or Google Cloud bucket, and then use the special custom field within aiWARE’s GraphQL API or CMS user interface to add the path to the state file.

https://medium.com/media/1ab90bbdd7c94a27a9d1f03cd2cb451e/href

Happy detecting!


Introducing Objectbox — super simple object detection was originally published in Machine Box on Medium, where people are continuing the conversation by highlighting and responding to this story.