Free tools and content for generating synthetic data
Check out our newly open sourced and academically released tools, datasets, and dataset generators for the creation of synthetic data for computer vision model training.
Unity Perception 1.0
The Perception package provides a toolkit for generating large-scale datasets for computer vision training and validation. Unity Perception 1.0 is a new, more complete release that includes new labels, randomizers, samples, and rendering capabilities.
Data analysis and visualization in Python
PySOLO tools is a new open-source python package that provides utilities to analyze and visualize data in the new SOLO format.
Now available as an open-source release, Unity SynthHomes is a 100,000-image dataset of synthetic home interiors and an associated dataset generator binary.
Now available for academic use only, Unity Synthetic Humans is a 3D person generator built from the ground up for human-centric computer vision.
“Wherever there is a requirement for data to drive machine learning, there is a role for synthetic data. Creating synthetic datasets in a virtual world means you can create millions of images very quickly compared to going out to the field and taking pictures.”
“At Ouva, our patient monitoring platform used Unity Computer Vision to generate synthetic data and reduced our month-long live data capture cycles to a week, while our dataset grew by 10X and model accuracies improved by 5 to 10%.”
Ouva’s simulated healthcare data platform harnesses the power of synthetic data to improve model performance by over 10%, reduce labeling costs by up to $40,000, create balanced datasets in hours instead of weeks, and reduce iteration cycles from weeks to days.
In this interview, learn how Boeing worked with Unity to generate over 100,000 synthetic images to better train the machine learning algorithms of its augmented reality (AR)-powered aircraft inspection application.
Gain insight into how Passio combines Unity’s synthetic data with real-world data to expand its datasets and speed up AI training for AI and augmented reality (AR) applications.
Learn how AI startup Neural Pocket used Unity Computer Vision to significantly reduce computer vision model development costs and time to deployment (from 24 weeks to 1 week).
Unlock data-driven AI development
Learn more about Unity Computer Vision, how to explore our sample datasets and generate your own sample datasets with our prebuilt environments.
Unlocking intelligent solutions in the home
Find out how our tools and services enable development of more capable computer vision applications for the home while mitigating roadblocks and challenges.
Getting started with 3D content for synthetic data
Synthetic data is powered by your library of 3D assets. Learn about sources and techniques for acquiring 3D content for common computer vision problems.
The factory of the future
Download our report to learn the vital role of computer vision, robotics simulation and real-time 3D technology to the future of manufacturing.
AI and machine learning, explained
Get up to speed on key terms in machine learning, computer vision, synthetic data and more.
Teaching robots to see with Unity
Empower your robots to accurately pick up an object without explicit knowledge of the object’s location. See how to collect synthetic data and train a deep learning model to predict the pose of a given object.
Train object detection model with synthetic data
Discover how you can generate a massive synthetic dataset to train your machine learning models.
Generate and analyze synthetic data at scale
Learn how to use tools from Unity to generate and analyze synthetic datasets with an illustrative example of object detection.
Myriad use cases enabled by synthetic data
Synthetic data is helping many organizations overcome the challenge of acquiring labeled data for training machine learning models. Discover the breadth of use cases it enables.
Can you find Waldo using synthetic data?
See how Unity’s Perception package was deployed to create Waldo-like images for training a neural net, which was then trained using the fastai library.
Create synthetic images for deep learning
Follow this tutorial to learn how to set up Unity and the Unity Perception package to create synthetic images that train neural nets in deep learning, AI and computer vision.
Synthetic aided computer vision algorithm development
See how Standard Cognition used Unity to reduce the financial costs and algorithm development time for data collection and labeling in their digital checkout system.
Frequently asked questions
Check out our papers to see how models trained with synthetic and real data outperform models trained using only real data:
Our customers use Unity to generate synthetic data for a variety of computer vision applications, including human detection, object detection, manufacturing defect detection, consumer electronics applications in the home, and more.
You can use synthetic training data when:
- You have only a small sample set of real-world data. In this case you can augment your real-world data with a large amount of synthetic data generated by Unity Computer Vision and boost your model performance.
- You are not able to collect the right real-world data for your project. In this case you can use Unity Computer Vision to generate high-quality labeled synthetic images and bootstrap your models with purely synthetic data.