The Task
A mining company needed annotated data to train a neural network that would automatically assess ore fragment sizes and detect oversized pieces in real time on a conveyor belt.
They had tried working with other vendors—but ran into trouble:
Validation was inconsistent. Internal QA took too long. They had to involve their own annotation team, which wasn’t sustainable.
They needed a reliable partner who could take over the entire cycle: annotation, quality control, and fast delivery—all within a week and a half.
To complicate things further, the ore images were collected in the field during a business trip. Delays weren’t an option.
The Solution
-
- 01
-
Fast Team Assembly & Workflow Optimization
- We formed a team of 13 annotators within 24 hours.
- From the start, we noticed a slowdown in the annotation software—each image contained a large number of polygons, and the system lagged.
Our workaround:
Split each image into four parts, annotate them separately, and then stitch everything back together for final review.
This allowed us to maintain speed without sacrificing precision.
-
- 02
-
Multi-Level Validation Process
- Each annotated batch went through several layers of QA.
- We streamlined communication between validators and annotators to reduce feedback loops.
- Our internal experts handled all edge cases—no client input was needed.
The Result
Fast Turnaround:
We completed the full cycle of annotation and validation in just 1.5 weeks.Reduced Overhead for the Client:
The client no longer had to manage annotation or QA internally.High-Quality Data:
The pilot confirmed data quality met production-grade standards, with no additional review needed from the client.
