Get started with synthetic data 

Training on synthetic data produces robust deep learning models without data collection, labeling, and data cleaning. This technique is the fastest way to bootstrap and improve computer vision models.

In this tutorial, you'll learn how to load an existing model trained on synthetic data and train a new model on 10,000 synthetic images.

Model trained on 100% synthetic data from 3D scans

Download the materials

Tell us a little about youself to gain instant access to all of the materials required to put synthetic data to the test.

You will be able to download:

  • code to load the model in a Google Colab
  • code to train the model in a Google Colab
  • 10,000 sample training dataset with COCO labels
  • pre-trained model showcased in the video above
  • validation dataset of 20 photos with COCO labels
  • SBX dataset manual for more advanced users

Thank you for sharing some info about yourself! Here are all of the materials you need to follow along with the tutorials.

These links will disappear when you reload the page:

Please don't hesitate to reach out to our team for help with the tutorials!

Oops! Something went wrong while submitting the form.

Follow along with our tutorials

Load & run a model trained on 100% synthetic data

In this video we load a pre-trained segmentation model and evaluate it on some test images. Follow along with our Colab Notebook!

Train a Mask R-CNN Object Detector on synthetic data

In this video train a new segmentation model (Mask R-CNN) on a small set of training images.

Download the full training set of 10,000 images and try it yourself!