Tools are emerging for managing, labeling, and visualizing datasets for machine learning, especially in computer vision (e.g., FiftyOne, Supervisely, Roboflow). Have any of you used these platforms in robotics applications? Given the importance of high-quality data, how useful do you find these tools in your workflow? Do they help improve model performance, or are there specific challenges you encounter when applying them to robotics?
Platforms like these support team-based workflows, allowing multiple users to annotate, label, and review datasets simultaneously, which can be useful for larger robotics projects involving multiple data sources. You could also touch on their integration with cloud-based environments, which can be beneficial for scaling dataset processing and model training.
Do you see these benefits along with reduced dev time offset the resultant costs and resource requirements?