Augmenting Datasets using Generative AI and Amazon Sagemaker for Autonomous Driving Use Cases on AWS

This article discusses how generative AI and computer vision techniques can be used to augment datasets for autonomous driving use cases on AWS. It highlights the challenges of capturing sufficient real-world driving data and proposes using generative AI models to create synthetic data representing various driving conditions.

Specifically, the article covers:

• Dataset augmentation using generative AI models like GANs and VAEs to modify weather conditions, add/remove vehicles, etc.
• A proposed workflow combining models like Grounding DINO, Segment Anything, and Stable Diffusion to manipulate images and videos based on text prompts
• Steps like object detection, semantic segmentation, inpainting, and optical flow to realistically modify scenes
• Examples of applying these techniques to replace road surfaces, change weather conditions, and introduce temporal consistency in videos
• Conclusion: While still a research topic, this approach could help create more robust autonomous driving models by exposing them to diverse synthetic data