Key Results

Centralized 2TB of visual data in one unified hub and eliminated scattered tooling
Improved stakeholder collaboration through shareable datasets that enable data-driven decisions.
Reduced false positives and detected edge cases by gaining clear visibility into where models struggle.

“During calibration for one of our 'follow-me' lighting features, FiftyOne revealed unexpected edge cases coming from window reflections, glare, and nonlinear wall proximity. It really helped us not only address this problem but also prove why a seemingly simple problem wasn’t so simple and required a more nuanced, data-driven approach.” — Chris Hall, Principal Machine Learning Engineer, Vivint

Introduction: Vivint

Vivint Smart Home is a leading provider of smart home technology that makes homes safer, smarter, and more energy efficient. With millions of cameras deployed across doorbell, outdoor, and indoor devices, Vivint processes massive volumes of visual data to power intelligent home security experiences. Vivint’s team of computer vision experts builds advanced models that detect people, vehicles, animals, and packages—delivering real-time insights and peace of mind to millions of homeowners.

Challenge: Lack of standardized data tooling and workflows

Vivint works with massive amounts of visual signals across images and videos. Before FiftyOne, data and annotations were scattered across various tools and with varying formats, making it harder to hand off projects or diagnose model behaviors. The lack of a consistent data infrastructure hampered collaboration and visibility. Challenges included:

Data sharing and visualization across images, video, bounding boxes, segmentation masks, and predictions.
Curation of balanced datasets that avoid bias, such as time of day, season, region, and camera firmware/model.
Annotation QA and standardized model evaluation across projects.
Picking up where a teammate left off without reinventing internal tooling for every initiative.

The team needed a solution that would prevent constant tool-building, standardize workflows, and make their massive datasets accessible and visual.

Why Vivint chose FiftyOne

The ML team at Vivint was drawn to FiftyOne's data-centric AI philosophy.

"Data-centric AI is the right approach for any organization working with visual AI. FiftyOne has led the charge in building a tool around that and has pushed Vivint toward truly adopting a data-centric approach. It has become the place where our team looks first to understand what’s working and what isn’t, before we iterate.” – Chris Hall, Principal Machine Learning Engineer, Vivint

FiftyOne became a tool of choice for the following reasons.

One platform for the entire ML team: Unlike the partial custom solutions engineers had to build, FiftyOne provided comprehensive features out of the box.
Data-centric approach: FiftyOne's focus on understanding and improving data quality aligned perfectly with Vivint's belief that better data, not just better models, drives accuracy improvements.
Team collaboration: The platform enables standardization, making all data accessible and shareable from one place, ending the cycle of reinventing data tools for every project.

Solution: A unified visual data pipeline for managing millions of camera deployments

Vivint implemented FiftyOne as its central visual data hub, consolidating approximately 2TB of camera data to create standardized ML workflows—from data curation to model evaluation and feature prototyping.

How Vivint uses FiftyOne

Data curation and QA: Vivint collects large volumes of data, with each camera sample carrying rich contextual information, such as timestamp, season, camera settings, firmware, model predictions, and location. With FiftyOne, the ML team visualizes the data and analyzes these distributions to prevent bias, select balanced subsets for labeling, and run QA on annotations.

Model evaluation and analysis: After training, the team evaluates model performance across lighting, weather, and camera types to pinpoint false positives and understand where models struggle–an important part of Vivint’s ML workflow.

“With FiftyOne, we know a lot better where our models have problems because we can look at false positives directly in the tool and link them to the underlying data samples.” — Chris Hall, Principal Machine Learning Engineer, Vivint

Object detection, tracking, and visual search: Object detection and tracking are important tasks for Vivint’s AI models, alongside other capabilities such as enabling notifications or offering interactive experiences like presence-aware ‘follow-me’ lights. The team leverages FiftyOne to detect and track human poses, objects, and packages. FiftyOne is also used to build searchable video tools using image and text-based search to instantly locate similar scenes through volumes of video data.

Results: ML efficiencies and solid team collaboration

Since implementing FiftyOne, Vivint has added efficiencies to their computer vision workflows and improved stakeholder alignment.

Efficient edge case discovery: With FiftyOne, the ML team was able to find interesting edge cases during development and drive continuous model improvements.

“During calibration for one of our 'follow-me' lighting features, FiftyOne revealed unexpected edge cases coming from window reflections, glare, and nonlinear wall proximity. It really helped us not only address this problem but also prove why a seemingly simple problem wasn’t so simple and required a more nuanced, data-driven approach.” — Chris Hall, Principal Machine Learning Engineer, Vivint

Improved collaboration and stakeholder communication: The ability to share filtered datasets via links has proven invaluable for their ML team. FiftyOne has enabled the ML team to easily align with stakeholders and make collaborative, data-driven decisions.

"I love that we can send a link to a data sample or even a filtered dataset to a stakeholder. They can look through it and observe the patterns that you're seeing. This level of collaboration has vastly improved our ML work." — Chris Hall, Principal Machine Learning Engineer, Vivint

By consolidating computer vision workflow in FiftyOne, Vivint Smart Home has built a scalable, efficient system for developing and deploying models that deliver intelligent experiences to millions of customers.

Talk to a computer vision expert