What is NVIDIA Omniverse™ Replicator?
NVIDIA Omniverse™ Replicator is a synthetic data generation framework that enables the generation of datasets used for AI model training in a virtual space built on the Omniverse platform.
Training an AI model requires a huge amount and variety of datasets. Collecting a sufficient dataset can be costly, time consuming, and difficult or dangerous in some locations. By using Omniverse Replicator, it is possible to automatically generate dataset images used for AI learning under various conditions such as object placement positions, angles, lighting, cameras, textures, etc. The generated dataset images are It supports multiple formats such as bounding Box, depth and segmentation.
Features of Omniverse Replicator
It is possible to import and use the photoreal virtual space created by Omniverse in the environment for creating the dataset. You can use the virtual space of the facility where the robot actually runs, such as a restaurant, office, outdoor, warehouse, etc., to build a shooting environment with various conditions. Since the camera angle can be set freely in the virtual space, it is possible to generate dataset images that reproduce various scenes.
restaurant
Macnica warehouse
Main function
Semantics Schema Editor: Adding label information
For the data set used for AI learning, it may be necessary to prepare training data that labels the objects in the scene for the original image of the scene. By using Semantics Schema Editor, it is possible to easily add the label information necessary for generating teacher data to objects in the virtual space.
■ Example
Specify the label name “cone” in the tab of Semantics Schema Editor
Synthetic Data Visualizer: Dataset Visualization
Various data formats are used for the training data of the dataset used for AI learning, such as bounding Box that indicate the coordinates of objects and segmentation that colors the areas where objects exist. With Synthetic Data Visualizer, it is possible to display objects labeled in virtual space in various data formats.
■ Example
Specifying bounding Box and segmentation as the data format will show both output image images
Synthetic Data Recorder: Dataset output
After reproducing the scene you want to shoot in the virtual space and confirming the image of the output dataset, the dataset image is generated by actually pairing the shot image and the training data. Using the Synthetic Data Recorder, you can set the number of captured images, etc., and automatically output the dataset image file.
■ Example
Generate dataset images by setting the number of captured images to 100
Random Parameters: Generating Scene Variations
With Omniverse Replicator, it is possible to randomly change various parameters that define the scene of the simulation environment in virtual space. It is possible to automatically generate a wide variety of data sets from a single shooting scene, including parameter variations such as object position, size, and color, as well as camera angles and lighting conditions.
Click here for detailed information
For more information on Omniverse, download materials are available.
Please use all means.
Contact Us
Macnica provides NVIDIA software solutions centered on Omniverse and has a rich track record of supporting robotics-related companies. If you are considering introducing AI, please feel free to contact us.
AI TRY NOW PROGRAM
This is a support program that allows you to test the latest AI solutions on the NVIDIA development environment before introducing them into your company.
You can deepen your understanding of software products such as NVIDIA AI Enterprise and NVIDIA Omniverse and investigate the feasibility of your implementation objectives in advance.
Related product page
