Parallel Domain Introduces Reactor, a Generative AI-Based Synthetic Data Generation Engine

Synthetic data platform parallel domain today announced the launch of Reactor, a next-generation synthetic data generation engine that integrates advanced generative artificial intelligence technologies with proprietary 3D simulation capabilities. The platform aims to provide machine learning (ML) developers with control and scalability, enabling them to generate fully annotated data that improves AI performance and fuels the creation of more secure and resilient AI systems for the world’s applications. real.

According to the company, Reactor improves AI performance in various industries, such as autonomous vehicles and drones, by generating high-quality images. Furthermore, the tool harnesses the power of generative AI to produce annotated data, which is a crucial requirement for ML tasks.

By generating bounding boxes (for object detection) and panoptic segmentation annotations (providing full/panoramic views), Reactor ensures that AI models can use visual data effectively, leading to more accurate and reliable results.

“Our proprietary generative AI technology enables users to create and manipulate synthetic data using intuitive natural language prompts while generating the corresponding labels required to train and test ML models,” Kevin McNamara, CEO and CEO, told VentureBeat. Founder of Parallel Domain. “Reactor’s ability to generate various synthetic examples has led to significant performance improvements in tasks such as pedestrian segmentation and debris and baby stroller detection. Its ability to improve the diversity of data sets, particularly for rare classes, contributes to superior model training.”

Refinement and rapid iteration of the ML model

The company said its tool allows users to create a wide range of synthetic data to train and test perception models. This is achieved by integrating Python and natural language, eliminating the need to create time-consuming custom assets and simplifying the workflow to improve efficiency. As a result, ML developers can quickly iterate and refine their models, reducing response time and accelerating AI development progress.

“Integrating these technologies into our platform allows users to generate data using Python and natural language commands, which improves the flexibility of synthetic data generation,” McNamara told VentureBeat. “Reactor equips ML developers with control and scalability, redefining the landscape of synthetic data generation. With Reactor, users can generate almost any asset in seconds using natural language prompts.”

Leverage generative AI to improve synthetic data pipelines

According to McNamara, while other companies use generative AI to create visually compelling data, it cannot be used to train ML models without annotations. Reactor overcomes this limitation by generating fully annotated data, which improves the ML process and allows developers to build safer and more effective AI systems.

“We leveraged generative AI and 3D simulation to create a wide range of realistic, detailed synthetic data,” McNamara told VentureBeat. “Generative AI enables the production of various scenarios and objects, while 3D simulation adds physical realism, ensuring the robustness of AI models trained on this data. Until now, generative models have had trouble understanding what they are generating, making them very poor at providing annotations like bounding boxes and panoptic segmentation, which are crucial for training and testing AI models.”

McNamara said the tool provides a wide spectrum of scene and data customization options. In addition, its adaptive background creation feature allows easy modification of generated scenes, allowing ML models to generalize across various environments. For example, users can transform a suburban California setting into a bustling downtown Tokyo scene.

Intuitive imaging

Reactor’s natural language cues introduce an intuitive way to generate image variations, according to McNamara. Users can modify existing images by simple prompts such as “make this image look like a blizzard” or “put raindrops on the lens.” This streamlined customization process eliminates the need to wait for custom asset creation, improving efficiency and response time.

“The adaptive background creation feature in Reactor enriches the diversity of training environments for ML models,” explained McNamara. “This broadens the scenarios the model can be trained on, helping it to better recognize and respond to different real-world conditions.”

Generative architecture allows models to understand the structure of generated objects and underlying scenes, making it easier to extract pixels and understand spatial semantics from layers in the generative process. This results in fully automatic and accurate annotations.

More diverse and realistic synthetic data

With Python, users can flexibly configure their synthetic data sets by selecting various parameters such as locations (San Francisco, Tokyo), environments (urban, suburban, highway), weather conditions, and agent distribution (pedestrians and vehicles).

Once the foundational dataset is configured, users can use Reactor to enhance their synthetic data with greater diversity and realism. Through the use of natural language cues, users can introduce a wide range of objects and settings into the scene, such as “garbage can”, “cardboard box full of sunglasses spilling on the floor”, “box of wood with oranges” or “carriage”.

Reactor generates synthetic data with essential annotations, including bounding boxes and panoptic segmentation, significantly speeding up ML model training and testing.

McNamara said the tool “revolutionizes” the traditional custom asset creation workflow, which typically involves a time-consuming design process, manual configuration, and integration by artists or developers.

“Rapid customization features powered by generative AI improve efficiency and improve response times,” added McNamara. “As a result, developers can create and integrate new assets into their synthetic data sets almost instantly, enabling faster iterations and continuous improvement of their models.”

Detailed visual information for autonomous vehicles

The company said it has seen notable improvements in the safety of autonomous vehicles and advanced driver assistance systems (ADAS) for cars. He also claimed that through advanced diffusion techniques, the tool recently achieved remarkable results in real-world scenarios.

In addition, the company noted that the tool recently significantly improved semantic segmentation results in the highly esteemed Cityscapes data set — a widely recognized benchmark for autonomous driving.

“Real-world data often lacks sufficient training examples for these less common but crucially important objects,” McNamara explained. “Reactor was used to generate synthetic data representing various scenarios involving strollers to close this gap. By feeding these synthetic data into training sets, models could better learn and generalize stroller detection to real-world scenarios, thus improving the safety of autonomous systems.”

He added that for the Cityscapes dataset, Reactor generated synthetic instances of trains and fed them into the dataset.

“These enriched data resulted in improved model performance in train detection and segmentation, which contributed to safer and more efficient autonomous driving systems,” McNamara said.

He added that several of Parallel Domain’s customers have recently begun incorporating Reactor’s capability into their AI development workflows. Although still in the early stages, the company is excited about Reactor’s potential to improve ML models.

“Both clients and the Parallel Domain ML team have trained models for cases that have significantly outperformed previous benchmark performance,” McNamara said. “This is because the variety of Reactor examples significantly increases the diversity of a data set. Diverse data trains great models and we are redefining the landscape of synthetic data generation.”