header object

How Keymakr helped
Cognex with OCR
and object detection
datasets

Industrial automation, machine vision

Company:
Services:
Overview:

Intro

Cognex is the world’s leading provider of vision systems, software, sensors, and industrial barcode readers used in manufacturing automation. Thanks to innovative industrial AI, Cognex technology improves product quality, prevents production errors, reduces manufacturing costs, and helps companies deliver high-quality products at an affordable price. The systems detect defects, monitor production lines, guide assembly robots, and track, sort, and identify parts.

Cognex works with many industries, including automotive, electronics, food and beverage, life sciences, logistics, OEM machine building, product security, vision-guided robotics, etc. The company uses data from billions of synthetic and real-world images to train its systems, which are then used to monitor operations, adjust production plans, and manage processes in real time across global supply chains.

Cognex also develops new technology. Its flagship In-Sight systems combine a camera, software, and processor in one compact industrial unit. It handles complex inspection tasks such as optical character recognition, assembly verification, and defect detection.

The challenge

Cognex approached Keymakr to label data for two main use cases: object detection and Optical Character Recognition (OCR).

Project challenges:

  • High accuracy. Any annotation errors directly affect production algorithms.
  • Data variety. Objects ranged from simple boxes to small parts and irregular items.
  • Flexibility. Work was conducted on demand, and sometimes data arrived in large batches that required urgent completion.
  • Scalable workload. Some tasks required rapid team scaling up and down to match the nature of the project.
The solution
1. Optical Character Recognition (OCR)

Cognex uses OCR to read text on a variety of objects, ranging from packages, labels, pharmaceuticals, tires, wafers, and electronics. Training and extracting characters help to identify and serialize products for which barcodes are not viable.

Keymakr created high-quality annotated datasets by:

  • Marking each character with a bounding box
  • Adding attributes for each character (letters or numbers)

Each bounding box included the symbol’s value to ensure accurate model training. While the task might seem simple, it was actually highly challenging and time-consuming. Many of the scenes contained numerous characters and overlapping elements, requiring Keymakr’s team to deliver extremely tight and precise bounding boxes. Maintaining this level of accuracy across such complex visuals was essential for achieving reliable model performance.

2. Object detection

Cognex has the capability of finding and isolating objects in typical scenes in the logistics industry. Keymakr worked to create the labels for locating and classifying packages and items on a variety of backgrounds, such as conveyor belts, cross-belts, tilt trays, and rollers.

By labeling tight, oriented bounding boxes to fit objects of any shape and orientation, Keymakr provided Cognex with accurate and reliable training data for their AI detection models.

Objects were classified into categories, for example:

  • Packages: cuboidal boxes of various sizes and orientations
  • Totes: plastic containers used for grouping and transporting loose or homogeneous items
  • Hazardous/irregular items (HAZR): unpackaged items, helmets, aerosols, irregular bottles
  • Other categories: jars, cans, bags, containers, or miscellaneous items like rolls, reels, and tapes

Depending on the scene, annotation difficulty could vary quite a bit. Some image sets could be labeled without any customer interaction, while others required a careful top-down analysis. Cognex and Keymakr worked closely and quickly together to set the expectations for labeling accuracy.

Collaboration features

Communication with Cognex was fast and efficient. The team received clear requirements, which helped them quickly train new members and meet even the most urgent deadlines.

Flexibility served as Keymakr’s added value. In the data annotation industry, there is often a tendency to outline rigid contracts and workflows, but Keymakr can provide “on-demand” data solutions. Batches of data are taken on as they arrive, and the team can pause and wait for the next assignment if none is available. This model enables Cognex to receive services precisely when needed, without being constrained by rigid schedules.

High execution speed, consistent annotation quality (confirmed at 99.53% accuracy), and flexibility with irregular tasks made the collaboration effective and reliable.

Results

Close collaboration: over a period of 1.5 years
Average annotation accuracy: 99.53%
Flexible delivery: Keymakr works “on-demand” while easily scaling the team.
Scalable operations: multiple OCR and object detection projects of varying sizes can run simultaneously.

This project demonstrated that high-precision OCR and object detection datasets can be produced at scale without compromising quality. Keymakr’s phased delivery, flexible workflow, and focus on accuracy ensured that Cognex could train reliable AI models efficiently.

"Keymakr has been an invaluable partner to Cognex by significantly reducing the time and effort required for data labeling thanks to their exceptional accuracy and reliability. By leveraging Keymakr’s services, our algorithm engineers have been able to focus on what matters most to Cognex - developing cutting-edge vision algorithms.

Their flexibility and precision have consistently exceeded our expectations. In fact, the quality of their labeling has reached a point where manual review on our end is virtually unnecessary. The process of creating a new project is seamless: upload a set of images, communicate expectations and accuracy requirements with the project manager, and let the labeling begin.

Throughout each project, we’ve been able to monitor progress using Keymakr’s intuitive online tools for image and label review. At the end of each sprint, they deliver well-structured, parsable JSON files that integrate effortlessly with our AI training pipelines.

We’ve been thoroughly impressed by Keymakr’s professionalism, responsiveness, and genuine care for us as customers. Their team is always ready to jump in at a moment’s notice to take on new image sets and keep the momentum going.

If you're new to AI development and considering doing labeling in-house, I strongly recommend investing in Keymakr. Their services are not only accurate and affordable but also a huge time-saver."

Nick Mullan, Cognex Engineer

"Cooperation with Cognex and specifically with Nick was always fast, efficient, and open. We maintain transparency regarding our estimations, timelines, and hours usage. There were no cases of any disagreement or misalignment. Our gratitude to the whole Cognex team."

Roman Gron, Keymakr PM

Reviews
on

down-line
high perfomer
high perfomer emea
leader
star
star
star
star
star

"Delivering Quality and Excellence"

The upside of working with Keymakr is their strategy to annotations. You are given a sample of work to correct before they begin on the big batches. This saves all parties time and...

star
star
star
star
star

"Great service, fair price"

Ability to accommodate different and not consistent workflows.
Ability to scale up as well as scale down.
All the data was in the custom format that...

star
star
star
star
star

"Awesome Labeling for ML"

I have worked with Keymakr for about 2 years on several segmentation tasks.
They always provide excellent edge alignment, consistency, and speed...

More
cases

down-line

Automotive

Delivering scalable traffic detection data for a major automotive company

MusicTech

Providing consensus-based media monitoring to help SoundAware protect the intellectual property rights of creators

AgriTech

How Keymakr supported Bluewhite in advancing real-world autonomy