- Platform
-
Services
Labeling Data
Processing Data
Improving ML Models
-
Who We Serve
For:•
ML Engineers
•
AI Business Executives
•
Product Managers
•
Academic Researchers
Technology & Innovation
Industry & Commerce
Academia & Services
- About
- Careers
-
Resources
- Pricing
- talk to sales
Vision AI
Pupil
Segmentation & Maritime Object Tracking
France
Overview
Nexvision needed detailed annotations across multiple vision tasks to support model training for gaze detection and maritime tracking.
(ellipse masks)
(bounding boxes)
(object tracking)
Client
French embedded vision company serving defense, aerospace, and industrial sectors.
Challenges
Needed high-precision labels across eye and maritime footage, with tool constraints and shifting batch volumes.
Solutions
Labeled 70k images and 4k videos using CVAT and client tools; 3 workflows ran in parallel with QA and feedback cycles.
Results
All datasets delivered on time; supported model training for gaze detection and maritime awareness with no quality concerns.
Client
Nexvision is a French company building advanced embedded vision systems for defense, aerospace, security, and industrial applications.
Challenges
Pupil detection required ellipse-level precision automated tools couldn't deliver
Maritime footage needed manual object tracking with dynamic camera motion
Tasks spanned different pipelines: some required internal tool, others CVAT
Batch sizes and project timelines changed frequently, requiring flexible delivery
Solution
Label Your Data split the work into three parallel workflows. Batches were delivered regularly, with feedback incorporated between cycles.
For pupil segmentation, 40 000 images were labeled using Nexvision’s internal tool with precise ellipse masks
For eye detection, 30 000 images were annotated with bounding boxes using CVAT
For maritime tracking, 4 000 videos were manually labeled with object tracks in CVAT
Training
Each stream began with written guidelines and a kickoff call.
Annotators completed small pilot batches before moving to full production.
No major rework was required thanks to strong initial alignment.
Results
3 complex annotation pipelines delivered in parallel
Gaze model training supported by 40k accurate ellipse masks
Maritime model improved with 4k hand-tracked video segments
Flexible team size adapted to shifting batch scope
All batches delivered on time with no quality issues reported
Start Free Pilot
fill up this form to send your pilot request
Thank you for contacting us!
We'll get back to you shortly
As a company developing optronic systems, we rely on high-quality annotated datasets to analyze sensor data. Label Your Data is fully integrated into our model-building process and has proven to be a flexible, reliable partner for our annotation needs.
Victor Martin-Lac
Machine Learning Engineer
Trusted by ML Professionals
Why Projects Choose Label Your Data
No Forced Commitment
Check our performance based on a free trial
Flexible Pricing
Pay per labeled object or per annotation hour
Tool-Agnostic
Working with every labeling tool, even your custom tools
Quality Backed by SLAs
We commit to accuracy and deadlines – or you don’t pay