VISIONPRO DEEP LEARNING

Solve challenging
applications with powerful deep
learning-based software

VisionPro® Deep Learning is an AI-based image analysis

software for complex applications. It automates tasks
that are too complicated and time-consuming to program
with rule-based algorithms, while providing a consistency
and speed that aren’t possible with manual inspection.
VisionPro Deep Learning tolerates natural variation
and differentiates between acceptable and unacceptable
anomalies to simplify development of highly variable
applications including:

▪ Defect detection and ▪ Material classification

segmentation ▪ Character reading,
▪ Assembly verification including distorted print
and part location

Simplified development Label checker

and deployment Simplifies image labeling by enabling
batch labeling and reducing the number
Expedites setup with fewer images and less
of images for review.
time required to train your application.

Common development environment Parameter autotune

Provides a scalable solution that enables Quickly train deep learning models for
you to leverage the latest technology without faster setup and deployment.
duplicating engineering costs.

Innovative toolset Faster line duplication

Addresses challenging applications that Accelerates expansion to multiple
require both high precision and high speed. production lines for easy scalability.
Set up a deep learning application with speed and precision
With a range of innovative features, VisionPro Deep Learning provides the capabilities and performance you need
at every stage of application development, from labeling and training to deployment.

LABELING
Label checker simplifies the labeling process, saving you time and effort. Run the checker to identify

1 images containing unmarked or mislabeled defects. Using this feature, you only need to review a
fraction of the image set – those flagged by the checker – compared to conventional labeling methods
that require you to manually label and review every image.

TRAINING

Parameter autotune calibrates deep learning models faster than conventional methods for quick
2 and easy setup. In one click, you can optimize your application, expediting training while ensuring
the accuracy of your results.

DEPLOYMENT

Faster line duplication accelerates validation and deployment of multiple production lines by allowing
3 you to adapt an existing deep learning model for a new line. This simplifies retraining and facilitates
additional deployments, enabling you to seamlessly ramp your capacity as your business grows.

VisionPro Deep Learning 2

Expedite deployment with intuitive graphical training
VisionPro Deep Learning’s graphical training interface simplifies the task of collecting images, training the neural
network, and testing it on a variety of image sets.

TOOL-CHAINING LABELING TRAINING

Unique tool-chaining Quick labeling Simplified training Parameter autotune

capability lets users and review process and re-training allows calibrates deep learning
break down their problem requires fewer images for easy deployment and models faster than
into smaller steps, and less time for setup. expansion to accommodate conventional methods.
making it easier to achieve multiple products and lines.
target performance.
2
3
1

VisionPro Deep Learning 3

Scale your operations with flexible common development environment
From low-level machine integration to building an application-specific HMI, VisionPro Deep Learning provides flexibility
in how you develop and deploy vision inspection in your production environment.
Tight integration with existing software and vision products creates greater compatibility across the Cognex product
continuum and allows you to introduce the latest technologies without duplicating engineering costs. It also enables you
to adapt an existing job to new or additional lines, delivering quick deployment into mass production environments.

Programmatic integration Graphical prototyping

Easily convert images, graphics, and results Integrate Deep Learning workspaces into Cognex
between VisionPro and VisionPro Deep Learning. Designer applications to simplify image acquisition,
results processing, and I/O2.

Fully deployable application Backwards and forwards compatibilities

Create and deploy VisionPro and Deep Learning meet every vision need
applications using Cognex Designer2. Train in the standalone Deep Learning Studio
or load a Deep Learning workspace into
Cognex Designer.

1
Only available in VisionPro Deep Learning 2.0

VisionPro Deep Learning 4

Automate your most demanding vision applications with innovative toolset
VisionPro Deep Learning is optimized for real-world industrial image analysis, requiring vastly smaller image sets
and shorter training and validation periods. Unlike traditional, rule-based tools, VisionPro Deep Learning tools are
trained by example. Choose between Blue Locate, Red Analyze, Green Classify, and Blue Read to deploy the
tools that’s right for your application.

Blue Locate for fixturing, counting, and assembly verification

The Blue Locate tool finds parts despite variations in perspective, orientation, luminance, glare, and color by learning
from the samples provided by the user. It detects features on noisy backgrounds, in poorly lit environments, on low
contrast parts, and even parts that flex or change shape.
Blue Locate is also a reliable solution for automating assembly verification. The tool can be trained to locate a variety of
components, even if they appear different or vary in size. By creating layouts based on the product being inspected, the
tool checks multiple feature locations and component types simultaneously, while adjusting to varying layouts.

Counts translucent objects Ensures kits are assembled correctly

Red Analyze for defect detection and segmentation

The Red Analyze tool finds subtle defects on a wide variety of part backgrounds and surface textures. It can be trained
to tolerate normal variations in lighting and part positioning, while detecting flaws, contamination and other defects.
For situations where it’s not practical to collect defect images, or where the defects are highly inconsistent, unsupervised
mode can be trained from good images only and identify cases that deviate from the normal part appearance. Red
Analyze can also be used to segment specific variable areas in an image to simplify other inspections.

Detects imperfections Isolates subtle defects

VisionPro Deep Learning 5

 Green Classify for object and scene classification

The Green Classify tool is a robust classifier that can be used to solve challenging classification tasks and distinguish
between different objects and defects. It identifies and sorts products into classes based on their common characteristics
such as color, texture, materials, packaging, and defect type. The tool tolerates natural deviation within the same class
and reliably distinguishes acceptable variation from different classes.

Classifies objects by type Classifies defects by type

Box Envelope Poly bag

Stain Dent Scratch

Blue Read for text and character reading

The Blue Read tool deciphers badly deformed, skewed, and poorly etched codes using optical character recognition
(OCR). Leveraging a deep learning-based, pre-trained font library, the tool can be trained to read application-specific
codes that traditional OCR tools are not able to decode. Plus, the visual debug feature identifies mis-reads that can
be easily corrected.

Reads embossed characters on

injection molded products Deciphers label-based codes on packaging

VisionPro Deep Learning 6

Select the best mode for your application needs
The Red Analyze and Green Classify tools can be used with two different network architecture settings:
Focused Mode or High Detail Mode. Focused Mode is ideal for applications where speed is paramount and
results are needed quickly. High Detail Mode is designed for more complex applications that demand maximum
accuracy over processing speed. Users can change between modes without re-labeling images and evaluate
which architecture best suits their application requirements.

Red High Detail Mode for Green High Detail Mode for
High-Precision Defect Segmentation High-Accuracy Classification
Accurate pixel-level defect segmentation is the primary Green High Detail Mode excels at identifying
advantage of Red High Detail Mode. Leveraging the subtle characteristics in an image or a region of
powerful and exhaustive High Detail Mode architecture, interest and classifying it accordingly. It provides
the tool can accurately learn the appearance of best-in-class classification accuracy even when
challenging defects and predict them in untrained dealing with significant variation. The tool is
images with pixel-level precision. Red High Detail Mode very versatile, tackling a range of applications
is a good fit for applications that call for both detection types from defect type classification to OK/NG
and measurement of challenging and subtle defects like classification, and part SKU classification.
blemishes, cracks, and scratches.

Up Down
Focused Mode High-Detail Mode

OK NG
Focused Mode High-Detail Mode

TensorRT
Enables higher speed processing
of high detail mode tools, without
sacrificing accuracy.

Batch Processing
Using the high detail mode for classification,
supports faster setup through parallel
processing of multiple images.

VisionPro Deep Learning 7

 SPECIFICATIONS
Windows based graphical user interface (GUI) with plugin support
Graphical & application
C library (Windows DLL) for runtime and/or training
programming interfaces
Microsoft .NET library (Wrapper for C library and WPF GUI components)
CPU Intel Core i7 or higher (recommended)
Cognex only supports NVIDIA GPUs
GPU Recommend GPU memory of 11GB or higher (GTX 1080Ti, RTX 2080Ti, 3070, 3080, 3090)
Note: VisionPro Deep Learning performance — in terms of processing time — will depend on hardware selection
Hardware & OS
Requirements RAM Memory 32 GB or more (recommended)
USB 1 free USB port (for the license dongle)
Windows 10 64-bit
OS
Windows Server 2019 64-bit
Storage Solid-state drive (SSD) with 100 GB or more of free space (recommended)
Supported image file formats PNG, BMP, TIFF, JPEG
Supported image properties 1–4 channels, 8 or 16 bits

Companies around the world rely on Cognex vision and barcode reading
solutions to optimize quality, drive down costs and control traceability.

Corporate Headquarters One Vision Drive Natick, MA 01760 USA

Regional Sales Offices Hungary +36 800 80291 Asia-Pacific

Ireland +353 21 421 7500 Australia +61 2 7202 6910
Americas © Copyright 2023, Cognex Corporation.
Italy +39 02 3057 8196 China +86 21 2279 9455
North America +1 844 999 2469 All information in this document is subject to
Netherlands +31 207 941 398 India +91 7305 040397
Brazil +55 11 4210 3919 change without notice. All Rights Reserved.
Poland +48 717 121 086 Indonesia +62 21 80602011
Mexico +800 733 4116 Cognex, In-Sight, and VisionPro are registered
Romania +40 741 041 272 Japan +81 3 5977 5400
trademarks of Cognex Corporation. Cognex
Europe Spain +34 93 299 28 14 Korea +82 2 539 9047
Designer is a trademark of Cognex Corporation.
Austria +43 800 28 16 32 Sweden +46 21 14 55 88 Malaysia +6019 916 5532
All other trademarks are property of their
Belgium +32 289 370 75 Switzerland +41 445 788 877 New Zealand +64 9 802 0555
respective owners.
Czechia +420 800 023 519 Turkey +90 216 900 1696 Singapore +65 3158 3322
France +33 1 76 54 93 18 United Kingdom +44 121 29 65 163 Taiwan +886 02 7703 2848 Lit. No. VProDL-DS-02-2023
Germany +49 721 958 8052 Thailand +66 6 3230 9998
Vietnam +84 98 2405167 www.cognex.com