Outsourcing of the design and development of AI systems
SETTING UP AND DEVELOPING NEURAL NETWORKS COMES WITH A HOST OF PROBLEMS:
Large datasets must be labelled. Up to 60-80% of project time is allocated to data collection and processing, training, and solution testing.
Highly paid professionals have to dedicate time to routine data collection and processing rather than spend it on product development
Developer companies may lack expertise in neural network training and testing
WHAT WE DO
WE SIMPLIFY THE WORK OF COMPANIES THAT DEVELOP ARTIFICIAL INTELLIGENCE SYSTEMS. WE DELIVER HIGH QUALITY SERVICES IN DATA COLLECTION AND LABELLING, NEURAL NETWORK TRAINING, AND SUPPORT OF OFF-THE-SHELF AI-BASED SOLUTIONS. WE HELP MAINTAIN THE DESIRED RATE OF DEVELOPMENT AND MAKE SURE THE PROJECT ACHIEVES ITS GOALS
Person 1Person 2Person 4Person 5Person 3
Our outsourcing services
Data collection
Collection of the required dataset (video, images, etc.) from open and unique sources
Consulting, planning of the creation, training and development of neural networks, testing of the available versions of neural networks
Five steps of a data labelling project
Analysing the peculiarities of project goals and objectives, drawing up terms of reference and negotiating a detailed work plan
Collecting data and preparing them for labelling
Data labelling
Testing the applicability of labelled data on intermediate results of neural network training
If necessary, adjusting the data labelling procedure or finishing work
EXAMPLE OF A COMPLETED DATA LABELLING PROJECT
Cameras scan the face and a food tray
A camera automatically recognizes the face and identifies the client
Peter Johnson
A camera automatically recognizes the face and identifies the client
Pancakes with berries 300 g
Orange juice 200 mL
The system recognizes food items and maps them to the menu
The system automatically withdraws money from the customer’s bank account (no cashier is involved)
END GOAL OF THE PROJECT
Develop a neural network-based system for real-time recognition of arbitrary sets of dishes on a tray with an accuracy of 98.5% or more.
LABELLING RESULTS
A photo shoot was organized on the basis of 5 corporate cafeterias at different companies
Over 400 Gb of HD-quality photos were collected
A total of 400 classes of labelled items were created
Over 1 million photos were labelled
COMPETITIVE ADVANTAGES OF SMS. AI LAB
Tasks of any scale can be completed rapidly thanks to the wide selection of professionals at our disposal and an open team approach.
A team of professionals with expertise in creating and developing AI and neural networks
A proprietary methodology and documentation for each stage of neural network development
A proprietary pool of data labelled under a variety of conditions
24/7 customer support
FEATURES OF THE DATA COLLECTION PROCESS
Deep expertise in the search and selection of prepared data banks
Professional operators collect photo data. We use special equipment that allows to shoot objects at different environmental, lighting and location conditions
Wide opportunities for organizing shooting at various types of objects
FEATURES OF THE DATA LABELLING PROCESS
Ability to label any kind of objects
Support for any methods of data labelling: semantic, polygon, block, linear, point
High quality standards and internal data labelling regulations
Data labelling speed: Photo – from 50 images / hour