Computer vision

How Softarex Develops Computer Vision-Based Systems

Computer vision-based software development requires the use of a wide range of technologies and approaches.

The development of any modern data mining system, including computer vision systems based on CRISP-DM (Cross-industry standard process for data mining) is divided into 6 main stages:

Business understanding (analysis of customer business need)

Data understanding (data visualization, feature exploration, target definition)

Data Preparation (data parsing, data preprocessing)

Modeling/optimization (preparation of model or optimization task -> training the model or performing task optimization -> preliminarily evaluation)

Evaluation (model evaluation, metrics computation)

Deployment (model deployment with the best metrics results)

Additionally, we use ASUM-DM (Analytics Solutions Unified Method, extension of CRISP-DM).

Approaches we apply to computer vision systems development include:

Decomposition of the image processing task into basic steps for video stream analysis;
Dataset preparation;
Exploratory data analysis (correlation analysis, cluster analysis, factor analysis, discriminant analysis, regression analysis);
Data visualization;
Feature selection / dimensionality reduction;
Automated model selection;
Automated hyperparameter optimization;
Multi-task learning;
Transfer learning.

Algorithms we use in computer vision-based software development:

Deep Learning: Convolutional Neural Network, Long Short-Term Memory, Deep Residual Networks, Convolutional Recurrent Neural Network, Fully Convolutional Network, Multi-task Cascaded Convolutional Networks, Region Proposal Network, Feature pyramid warping, ArcFace Network
Decision Tree, Naive Bayes Classifier, Linear Regression, Bayesian Networks
Logistic Regression, Support Vector Machine, Random Forests, Gradient Boosted Trees
K-means, Fuzzy C-means, Gaussian mixture models
DBSCAN, EM-algorithm, Kohonen Neural Network

AI, Computer vision and Data science

OpenCV, DLib, Tensor flow, CUDA, DeepDetect, OpenNLP, Storm, Spark, Hadoop, SystemML, Stanford Classifier, SmileMiner, DirectX, OpenGL, QT, Cognitive services from Amazon and Microsoft Azure

Extensive experience in designing and implementation of industrial solutions with Computer Vision
Profound knowledge of applied mathematics
Employing PhDs in applied mathematics
Ability to implement Computer Vision in different domains and capabilities to understand tasks from different domains
Capability to develop new Computer Vision algorithms for solving unique problems

Meet our
experts

Meet our experts 1/2

Maksim

System Architect and Delivery Manager

In over 10 years of my professional career, I have developed, architected, and managed dozens of projects for the Manufacturing & Energy, Restaurant & Hospitality, Aviation, and IT industries. I have hands-on experience with Mobile technology, Computer vision, Machine learning, Predictive modeling, Cloud computing, and SaaS. I managed the development of a Computer vision-based kitchen management system for a large restaurant chain. We did a great job on this project, and I ‌consider it the highlight of my career.

Yauheni

Lead Data Science and Computer Vision Engineer

With MA and PhD degrees coupled by multi-year scientific research experience, I am responsible for the development of knowledge-intensive solutions based on Computer vision, Predictive modeling, Deep learning and Machine learning technologies. My responsibilities include planning and managing research, designing software architecture, developing project technical documentation and integrating research results into final products. My team and I always strive to apply the best practices of Computer vision and Machine learning to projects.