Slide1.png

Flexible image-centric  AI/ML
engineering platform

compliant with Industry 4.0 standards
with strong

data management capabilities.

BeeYard supports ML/AI engineers and data scientists throughout the entire MLOps workflow and enables collaboration of all stakeholders within enterprises.
 

An ultimate solution addressing industrial-grade applications, BeeYard was designed to utilize the maximum amount of variable data from production, including images from industrial optic systems, to build robust ML/AI modules quickly and reliably.

 

BUILD ROBUST ML/AI MODELS FASTER

BeeYard_presentation_03.22_final-01.png

DESIGNED FOR INDUSTRIAL-GRADE
MACHINE VISION APPLICATIONS

END-TO-END ML/DS Platform

BeeYard covers the entire MLOps workflow:

  • data collection and organization

  • data preparation and pre-processing (annotations, labeling)

  • model training

  • deployment

  • maintenance

1.

2.

3.

4.

5.

beeyard prvky-01.png
beeyard prvky-02.png
beeyard prvky-03.png
beeyard prvky-04.png
beeyard prvky-05.png

Feasibility
Study / POC

Prototype

Deployment & Commissioning

Production

Maintenance

COVER THE ENTIRE MLOps WITH ONE
POWERFUL PLATFORM

beeyard prvky-06.png

Data cell
 

  • A set of all object - related data

  • Basic database unit

  • Context - driven data storage

  • bound to each object

icony-10.png
icony-09.png
icony-08.png
icony-05.png
icony-06.png


 

  • Created time

  • Modified time

  • Description

  • Tags

  • Properties

icony-11.png

DESIGNED FOR MACHINE VISION

  • large data volumes processing;

  • data collection and storage optimized for ML/AI tasks;

  • contextual data storage - DATA CELL APPROACH
    images are tight with context metadata from other sources;

  • hive database storage local or cloud.

INDUSTRIAL MACHINE VISION REQUIRES SPECIFIC
APPROACH TO DATA PROCESSING AND MANAGEMENT

SDKs

AUTOMATED

BeeYard leverages automation features to
accelerate ML/AI systems development:

  • AI-assisted labeling.

  • Data cleanups

  • Data preparations

  • Pretrained models for principal DL tasks,
    facilitate finetuning of these models.

ACCELERATE THE MLOps WORKFLOW AND
GET RID OF TIME-CONSUMING MANUAL TASKS

icony.png

BeeYard is designed to be embedded into CI/CD pipeline.

Polygon selections

AI assisted
labeling

Label classes

Key
points

Bounding boxes

Semantic
segmentation

DATA PREPARATION

Annotation jobs

  • all types of annotations required in industrial field

  • coordination between annotators

  • can be distributed to annotators anywhere with
    access rights management.

Image preprocessing
Image augmentation

  • images and annotations are already in the
    correct format to be feed to whatever computer
    vision algorithm or deep learning model,

Batch operations on images:

  • eliminate repetitive annotations by applying it to
    a batch of images in one shot.


THE QUALITY OF MACHINE VISION ALGORITHMS STARTS WITH VALID AND CONSISTENT DATASETS

COMPOSITE AI CAPABILITY

BeeYard supports deep learning as well as other AI
techniques (traditional machine learning, image
processing, or even a combination):

  • A combination of conventional machine learning
    and deep learning achieves the highest possible
    performance.

  • Deep learning is not transparent enough to be
    applied to all industrial applications (not possible to
    explain the outcomes and comply with e.g. quality
    regulations).

  • Teams can test both in parallel or use hybrid
    combinations (machine learning + deep learning).

DEEP LEARNING IS NOT A SILVER BULLET,
ESPECIALLY IN MANUFACTURING

a3_sklad_letak_beeyard.png
icony-21.png

BeeYard local storage

icony-15.png

Public Cloud

icony-17.png
icony-16.png
icony-18.png
icony-15.png

Private Cloud

icony-19.png
icony-20.png

HYBRID CLOUD SOLUTION

BeeYard supports hybrid scenarios
of data collection and storage

  • Edge *

  • On premise

  • Public or private Cloud

  • Or a combination.

BE COMPLIANT WITH INTERNAL CONSTRAINTS AND
POLICIES ON DATA GOVERNANCE AND SECURITY

COLLABORATIVE CAPABILITIES

Collaboration of all involved parties, team members,
3rd parties at all stages of machine vision projects:

  • audit logs

  • versioning

  • access rights

Collaborative annotations

BeeYard serves a communication tool for the dialogue between domain experts (in-house teams) and data scientists (inhouse/3rd party) party), typically quality managers who can label or annotate pictures directly in the platform.

INVOLVE DIFFERENT USERS TO PARTICIPATE
ON THE MACHINE VISION PROJECTS

BYSCHEMA DATA CELL-01.png

SINGLE POINT OF TRUTH
ONE DATABASE ~ MANY USERS

BEEYARD architecture.png

SCALABLE ARCHITECTURE

Deploy BeeYard from the local edge up
to enterprise level utilizing cloud storage:

  • EDGE level

  • FACTORY level

  • ENTERPRISE level

* Optional BeeYard EDGE device for the
collection of data from the edge devices
as part of the BeeYard product family.

SCALABLE ARCHITECTURE ENABLES TO PROPAGATE DATA
FROM THE EDGE UP TO THE ENTERPRISE LEVEL FOR FURTHER
PROCESSING OR ANALYSIS

DATA MANAGEMENT & SECURITY

BeeYard supports deep learning as well as other AI
techniques (traditional machine learning, image
processing, or even a combination):

  • ACCESS RIGHTS

  • USER MANAGEMENT

  • AUDIT LOGS

  • VERSIONS

  • Protection against the loss of data,

  • Automated backups,

  • Synchronization between multiple devices,

  • Easy data transfer,

  • Restore capabilities,

  • Security compliance.

SECURITY:

MANAGE WHO CAN MANIPULATE WITH YOUR DATA AND IN WHAT WAY, ESPECIALLY WHEN A WIDE THE AUDIENCE HAS ACCESS RIGHTS

icony-22.png
icony-23.png
DATAAGGREGATION-01.png

DATA MINING

Easy access to the correct and specific data:

  • easy aggregations,

  • queries management,

  • insights on data,

  • tools to output statistics of the data,

  • automated tests on specific batches of data,

  • possibility to add properties to data such as sensor
    measurements, date and time, dimensions etc.

EASY ACCESS TO RELEVANT PRODUCTION DATA FROM VARIOUS SOURCES

Proposed BeeYard Architecture
OPEN-SOURCE EXAMPLE

INDUSTRIAL-GRADE LIBRARIES EXAMPLE

Try BeeYard for your 
next machine learning
project