Synthetic data: The secret sauce behind our license plate readers
In the April issue of our newsletter, we introduced NVIDIA’s Omniverse platform. Omniverse enables artists, designers, researchers, and engineers to collaborate and create. Together, they can build and launch metaverse applications, intricate 3D worlds, custom extensions, and more.
This issue, we’re showcasing the first Omniverse extension that we built.
Introducing the license plate synthetic data generator (LP-SDG)
In the past, if we wanted to train our machine learning models on reading license plates, we’d have to collect lots of license plate photos. Then, we’d have to engage people to clean the data and annotate every license plate photo.
Omniverse was a game-changer because it allowed us to create the license plate data that we needed. This artificially generated data is also known as synthetic data. We developed the Automatic License Plate Reader (ALPR) datasets, which include all the data necessary for training our license plate reader models. These include license plates, plate numbers, bounding boxes, and more.
That’s not all. Omniverse enables us to create license plates under different conditions by tweaking the time of day, fogginess, lighting, and other environmental elements. We can even add some “damage” to the license plates to give them a worn-out look.
LP-SDG in action
Now, we can train our machine learning models on annotated data that we generated ourselves, and deploy products like PlateReader faster.
PlateReader is a device that gives users accurate and efficient real-time automatic license plate recognition. This is useful for parking management and monitoring the traffic in cities. Here’s what the PlateReader user interface looks like:
What’s next for LP-SDG?
We’re using LP-SDG in our vehicle playground project. This is a vehicle manipulation tool for users to modify LP-SDG parameters in real-time, according to their exact configuration preferences. That way, users don’t have to create full datasets and will have easier parameter visualization when building their own machine learning models.
After the vehicle playground is ready, we’ll be sure to share it on Omniverse. Meanwhile, if you want to try LP-SDG, download Omniverse. You can find the SmartCow LP-SDG Omniverse Extension under the extensions tab.