How it works

Market intelligence & Monitoring


Training data How it Works​

Training Data

How it Works Computer vision

Computer Vision

How it Works Data fusion

Data Fusion

Interested in custom datasets for your market intelligence?

We use a combination of Computer Vision and human-led verification in building datasets to assure the quality of collected data using airborne or satellite imagery. Context matter and Machine Learning only works with a good training data verified to be a ground truth.  

In a number of cycles, and with preexisting Computer Vision models data can be collected and verified using both imagery products and Map Tile Services.

Commonly Asked Questions:

How much data is required for AI training set?

We start from 1000 images per class, but we work closely with clients’ Machine Learning Engineers teams through the process of training, validation and testing to achieve high accuracy ML models

How long does it take to annotate AI training set?

Typically we can return the annotation within 2-3 days from submission, and set up
the workflow that fit your active learning process

What is the process of datasets development?

We customize the process individually, but typically with data fusion of web sources, imagery and maps.

How to ensure accuracy of deep learning solution for dataset?

We develop computer vision model through an iterative process and large validated dataset using supervised learning. But it is never an algorithm, it is always a verified data.