How a healthcare tech leader achieves 50x faster data curation

Introduction

This Fortune Global 500 healthcare technology leader in cardiac care, diabetes management, surgical robotics, and patient monitoring has transformed the lives of over 63 million people suffering from chronic diseases and acute conditions through medical devices and therapies. The company uses FiftyOne to improve production efficiency for manufacturing defect detection and anomaly analysis use cases.

The company's AI and data-science teams drive intelligent manufacturing automation using visual data to improve quality control and defect detection. Their AI for Manufacturing Imaging team builds infrastructure that enables teams worldwide to develop, deploy, and manage computer-vision models. These initiatives make AI development simple, scalable, and accessible for data scientists, software engineers, and factory-floor process engineers.

"With FiftyOne, we were able to cluster 45,000 images from production for an analysis, and instantly found anomalies: products stacked on top of each other, unexpected artifacts, etc. This would've taken days to detect manually." — Principal AI Engineer/Data Scientist

Challenge: Building a Scalable Data Foundation for Manufacturing AI

The company's manufacturing operations involve numerous product lines, imaging systems, and inspection processes, each generating massive volumes of visual data. The AI teams needed a way to harness these data streams for defect detection, anomaly analysis, and process optimization.

When the manufacturing AI team first set out to build their AI infrastructure, they attempted to develop a custom data management platform from scratch. The team faced challenges in scalability, flexibility, and collaboration, particularly:

Managing multiple manufacturing processes, each with unique defect types requiring classification, bounding boxes, or segmentation annotations.
Integrating multiple tools flexibly for annotation, model training, and evaluation.
Enabling easy collaboration between data scientists, manufacturing engineers, and R&D teams.
Exploring and managing visual data with fast rendering and analysis capabilities

They quickly discovered the immense complexity of creating a production-grade solution that scaled with their massive data. The team pivoted to finding a market solution and discovered FiftyOne.

Why FiftyOne was the tool of choice

The company chose Voxel51 as its visual AI platform primarily because of FiftyOne's data-centric management and plugin-based extensibility.

"Voxel51 sits at the center of our ecosystem. It's our jumping-off point for everything: from data management to annotation to model training. The plugin ecosystem has made it easy for us to integrate the tools we need, without building everything from scratch." — Principal AI Engineer/Data Scientist

The company selected FiftyOne for its:

Data-centric focus: built around data quality, exploration, and visualization rather than model-first workflows.

Flexible plugin architecture: FiftyOne's extensibility and interoperability provided the company with the flexibility to rapidly build custom integrations and workflows. Given the flexibility, the team could continue to incorporate their chosen ML tools for model training and experimentation.

User accessibility: The platform supports both technical and non-technical users (data scientists, engineers, process owners) in exploring and curating data.

Solution: Central data hub for manufacturing defect detection

The company implemented FiftyOne as the central data platform, where all projects begin and where the team stores and evaluates results from other tools.

Manufacturing defect detection and anomaly discovery: The team manages 36+ active projects spanning manufacturing processes across the company's facilities. In each project, the team accumulates images, organizes defects and labels (classification, bounding boxes, segmentations), and filters subsets for model training and feasibility assessment.

Visual clustering and exploration: FiftyOne's visual embeddings, patch-level analysis, and filtering capabilities help the company's teams explore data at scale. The platform enables them to select balanced subsets for model training and detect distribution gaps early in development. Discovering critical issues in development helps the team create robust test sets that account for all variations and defect types. These discoveries also enable them to provide valuable feedback to process engineers in production.

Results: 50x faster data curation for gathering manufacturing data insights

Since implementing FiftyOne, the team has transformed how they develop, evaluate, and deploy AI models for manufacturing quality control.

Faster data curation: When processing production data, the team downloads and samples from over 2TB of manufacturing imagery. Using FiftyOne's clustering capabilities, they were able to analyze 45,000-image datasets in minutes rather than days of manual review.

"With FiftyOne, we were able to cluster 45,000 images from production for an analysis, and instantly found anomalies: products stacked on top of each other, unexpected artifacts, etc. This would've taken days to detect manually. The value that FiftyOne provides is priceless!" — Principal AI Engineer/Data Scientist

Scalable foundation: The company currently manages 36+ active projects and plans to scale to domain-specific models trained on 1M+ images. FiftyOne provides the performance and flexibility they need to grow their AI capabilities across the organization.

Unlock FiftyOne extensible workflows

Discover how you can use state-of-the-art models and powerful add-ons to extend and customize FiftyOne functionality. Browse the curated collection of plugins or develop your own.

Learn about FiftyOne plugins ecosystem →