Automated Data Labeling Delivers
Rapid Results for Edge Vision

Edge Vision, a company specializing in AI for edge computing environments in smart cities, faced the challenge of collecting and building high-quality datasets for their machine learning models. These models are especially demanding, as they need to function at a high level of accuracy without access to the internet or additional resources. Previously, Edge Vision relied on third-party labeling services and a lengthy in-house QA and audit process, which consumed valuable time better spent analyzing business requirements, transforming them into technical tasks, and implementing them.

Before partnering with Superb AI, the team used a self-hosted instance of CVAT, which proved time-consuming and inefficient. All labeling and management work, including data uploading, task organization, and assignment to third-party labelers, was done through this tool. The need for Python scripts for post-processing complex data further slowed down operations. As Edge Vision scaled up its solutions, they realized they needed to find a more efficient way to process and annotate data without increasing their labeling budget, leading them to seek an alternative to manual labeling.

That’s when Edge Vision turned to Superb AI’s platform and Custom Auto-Label technology to quickly label objects like cars and pedestrians in complex, dynamic scenes. For more stuff-like object categories, including complex and irregular shapes, they also adopted Auto-Edit, an AI-assisted annotation tool. Auto-Edit allows for automatic segmentation by drawing a bounding box around the target and making revisions with a single click.

This technology enabled Edge Vision to reallocate a significant portion of their development sprints from labeling project management and QA to staying updated on the latest research, testing new algorithms, training models, and performing other data and model-related tasks. Adopting Superb AI's platform resulted in a more efficient data pipeline and a dedicated platform for improving training data and models.

In terms of results, these technologies immediately reduced annotation and QA time by 300% while decreasing labeling costs. Edge Vision’s ML team can now label new datasets in as little as an hour, with less than a week needed for curation, sampling, and iteration. The platform also improved communication and collaboration within the ML team when manual tasks were required, eliminating the need for the daily back-and-forth with external labelers. Using Superb AI has empowered Edge Vision to overcome dataset labeling challenges, enabling them to focus on developing and scaling more cutting-edge AI solutions for smart cities.