Google AI unveils a scalable dataset generator Kubric; know what it is
Google AI researchers demonstrated the effectiveness of Kubric in a detailed paper using a series of 13 separate datasets generated for tasks involving unsupervised multi-object video detection.
Researchers Klaus Greff, Francois Belletti of Google AI, Lucas Beyer of Google Scholar, and others have published a paper on Kubric, a scalable dataset generator.
An open-source Python framework generates high-quality images with PyBullet and Blender. PyBullet is used to train the model to interact with other objects physically, while Blender is used to render the images. The tool was created to reduce the costs and resources associated with producing mature and unbiased real data.
The research paper demonstrated Kubric’s effectiveness by employing a series of 13 distinct datasets generated for tasks involving unsupervised multi-object video detection.
The datasets were for various tasks, including 3D NeRF models and optical flow estimation. Kubric released photo-realistic scenes that are heavily annotated and can be easily scaled for larger tasks that require thousands of machines to complete. The tool can generate massive amounts of such synthetic data.
Despite the urgent need for less expensive, well-annotated, and unbiased data, there is a scarcity of software tools that produce effective, usable data.
Synthetic data has recently become more popular due to its numerous benefits, including lower costs, rich annotations, giving researchers complete control over their data, and avoiding risks associated with licencing and privacy.
For more such content, visit: https://bit.ly/2XkTP0P