A global Fortune 500 leader in automotive systems and industrial automation leverages computer vision and machine learning to drive safety-critical innovations across its business units.

In the automotive sector, the company develops a wide range of AI-driven perception systems, including driver and occupant monitoring, in-cabin safety, fleet operations, and automated quality inspection for manufacturing. These systems help reduce safety risks, improve regulatory compliance, and deliver production-grade reliability for consumers and OEM partners.

“FiftyOne is a critical data curation and model development tool in our AI projects. When we onboarded it, we had no idea how popular it would become." – Afonso, Machine Learning Operations Architect

Challenge: Gaining insights from petabytes of safety-critical visual data

The company’s automotive AI division develops and refines machine learning models for interior perception and safety systems, including:

Understanding occupant presence, position, and posture
Detecting unsafe passenger behaviors or states
Recognizing the correct installation and occupancy of child seats

The team manages massive volumes of visual data, 3-4 petabytes, from large-scale recording campaigns across multiple projects. Before adopting FiftyOne, data inspection and curation were handled through custom in-house tools that were increasingly difficult to maintain and scale.

Key challenges included:

Slow visual inspection: Reviewing large datasets with complex overlays was time-consuming and inefficient.

Difficult data management: The absence of a centralized, version-controlled metadata layer led to inconsistencies across datasets and models.

Slow model iteration: Fragmented tools and interfaces made it harder to enforce a unified data model throughout the MLOps lifecycle.

“Visual inspection was a big challenge for us for quite some time. We needed to quickly look at images with labels, bounding boxes, and poses overlaid. It wasn't practical to go through individual files; it just would not work at our scale, so we tried doing custom dashboards and web applications, but it was really hard to maintain and not very user-friendly. FiftyOne solved that need nicely, and it also completely covered our data curation and analysis needs.” — Afonso, Machine Learning Operations Architect

Solution: Scalable visual data management

FiftyOne was initially introduced to provide ‘insight at a glance’—helping teams quickly visualize large datasets and identify outliers in seconds. Its adoption quickly expanded as teams recognized the broader benefits of AI data management and collaboration.

Today, the company uses FiftyOne as a core visual data management and curation platform across multiple AI workflows, processing images, video, and 3D data with complex annotations such as bounding boxes, keypoints, and poses.

Integrated into the company’s MLOps ecosystem, FiftyOne enables:

Automated data ingestion: Custom pipelines automatically load new datasets into FiftyOne as recording campaigns complete.
Annotation workflows: Teams identify annotation mistakes and correct them directly within the interface.
Data discovery and curation: Engineers query, filter, and visualize scenes to identify data gaps and improve model coverage. Engineers explore and query datasets, understand scenes, and identify data gaps using FiftyOne.

“Over time, we began relying on FiftyOnemore and more, to the point that, today, it acts as our "gold layer" for ML workflows and nicely feeds into our Azure ML pipelines. Our AI Engineers are really happy with it; they can focus on improving the models instead of scripting ETL pipelines or trying to create other workflows.

There is a famous paper by two researchers from Google stating that mature ML systems typically have “(at most) 5% machine learning code”. Our experience with Voxel51 is that its extensibility and flexibility help our teams a lot with the remaining 95%: data preparation is simpler; the API and SDK are really good; it fits our infrastructure constraints and integrates nicely with popular third-party tools.” — Afonso, Machine Learning Operations Architect

FiftyOne: A critical tool for visual perception use cases

FiftyOne is a mission-critical part of the company's AI development workflow. With full team adoption, the organization eliminated data synchronization bottlenecks and transformed how teams collaborate on datasets. FiftyOne now drives faster review cycles, cleaner datasets, and more accurate perception models.

Voxel51 is mission-critical to our AI development. Without it, our visual perception projects would be much slower. Voxel51’s data curation and analysis capabilities and team collaboration have become essential to how we develop AI systems." — Afonso, Machine Learning Operations Architect

Talk to a computer vision expert

Challenge: Gaining insights from petabytes of safety-critical visual data

Solution: Scalable visual data management

FiftyOne: A critical tool for visual perception use cases

Talk to a computer vision expert