Image Classification
Cutting-edge Image Classification solution that leverages the unique capabilities of the Intel Neural Compute Stick 2
This case study highlights how Onix ensured integration that combines the power of AI with the advantages of edge computing, bolstered privacy measures, and access to Intel's robust AI ecosystem.
Automatic Electric Drive
Industry
10 specialists
Team size
Kyiv, Ukraine
Location
2022 – 2023
Project duration
[ Business Context ]
In the contemporary landscape, industrial enterprises encounter many intricate image classification challenges that significantly influence their operations.
These challenges encompass a broad spectrum of concerns, such as defect detection, quality control, component recognition, anomaly detection, efficient inventory management, verifying the accuracy of barcodes, labels, packaging, and many more.
These multifaceted image classification challenges directly impact operational efficiency, product quality, and safety within industrial enterprises.
Recognizing the significance of these issues, our client approached us with the visionary idea of developing a comprehensive image classification solution. The aim is to harness the power of advanced technology to effectively address these challenges, thereby enhancing productivity, ensuring safety, and optimizing their industrial processes.
[ Product scope ]
The responsibilities of the Onix team included:
Collecting requirements to establish project objectives and define the project scope.
Ensuring a development process that is efficient and transparent.
Developing software that offers real-time image classification and analysis, eliminating delays experienced with cloud-based solutions.
Integrating Intel Neural Compute Stick into an image classification pipeline, ensuring an app runs sophisticated AI models directly on user devices without cloud processing.
Incorporating dependable and cutting-edge technologies to guarantee the software's high performance and seamless functionality.
Have a project in mind? Let's talk!
The Onix team crafted a reliable solution that benefits five user groups:
01
Manufacturers
For those in manufacturing, quality control is paramount. The software can help identify product defects or anomalies on the assembly line in real time. This reduces wastage and ensures product quality.
02
Healthcare professionals
Doctors or radiologists can use the tool to quickly analyze medical images, assisting in early diagnosis or anomaly detection.
03
Wildlife researchers and conservationists
Researchers can track animal movements, numbers, and behaviors more efficiently by classifying species from trap camera images.
04
General public
For individuals keen on comprehending and organizing images for educational, research, or recreational endeavors.
05
Retail
Retailers can classify products, track inventory visually, or even help customers find products with similar attributes.
Solutions we provide
Real-time edge classification with NCS2
Instead of relying on cloud servers or powerful local machines for image processing, the app leverages the power of the NCS2 for on-the-spot, real-time image classification. This ensures speedy results with minimal latency, making it ideal for situations that demand immediate feedback.
Our team integrated the Intel NCS2 with the core application, optimizing the machine-learning models with the OpenVINO toolkit. This ensures the models run efficiently on the NCS2, providing users instantaneous classifications. The backend seamlessly detects the presence of the NCS2 and reroutes image processing tasks directly to it.
Model optimization
Our specialists used the OpenVINO toolkit to optimize and adapt the machine learning models specifically for the NCS2. Multiple iterations ensured the models remained accurate and efficient.
Adaptive learning with user feedback
The application is not just static; it learns and adapts. If a user disagrees with a classification, they can provide feedback, and over time, this feedback is used to refine and improve the classification model.
Our team created an intuitive feedback mechanism in the UI, allowing users to flag incorrect classifications quickly.
Business goals we helped to achieve:
Real-time image analysis
The software offers real-time image classification and analysis, eliminating delays commonly experienced with cloud-based solutions.
Data privacy and security
With edge computing capabilities, businesses can assure customers that their images are processed locally and aren't uploaded or stored in the cloud, thereby addressing data privacy concerns.
Cost-effective solution
Businesses can reduce infrastructure expenses tied to maintaining large cloud-based processing servers.
Scalability
Since the computational load is distributed and not centralized, businesses can easily scale the solution to more users without significant backend enhancements.
Core technology stack we used
Python,
OpenVINO,
MobileNet,
TensorFlow,
FastAPI,
Google Colab
Results
Our client now possesses a finished product that stands as a testament to innovation and efficiency. The outcome is a versatile, robust, and exceptionally user-friendly image classification application meticulously optimized for seamless integration with the Intel Neural Compute Stick 2.
The Onix team developed multiple deep-learning models optimized specifically for some use-case scenarios. These models were trained on high-quality datasets and then pruned and optimized using Intel's OpenVINO toolkit to ensure they run efficiently on the NCS2.
The combination of edge computing through the Intel NCS2, real-time image classification, and an emphasis on user privacy and data security makes this software solution invaluable for businesses and end-users looking for quick, reliable, and secure image analysis.
See other related projects
Kazakhstan
Mobile Communications
Centralized aggregator classifying news
into 10 predefined topics for 40K users
Optimizing news access with LSTM categorization
Centralized aggregator classifying news into 10 predefined topics for 40K users
Services provided:
Onix provided solution for streamlined news consumption, language detection, database development, intuitive UI, and services integration
Australia, Sydney
Entertainment
An animation software to ensure
seamless, comedic & realistic face
replacement
AI-based solution to replace faces in video stream
An animation software to ensure seamless, comedic & realistic face replacement
Services provided:
We crafted an AI-driven solution to replace faces in humorous videos, utilizing advanced face detection, dynamic object mapping & accessory augmentation
USA, Florida
Beauty, Cosmetics
AI-powered iOS app that allows
consumers to easily find safe beauty
products
AI-powered iOS app for cosmetics & beauty industry
AI-powered iOS app that allows consumers to easily find safe beauty products
Services provided:
We developed an iOS skincare app with an intuitive admin panel, robust security, AI-driven recommendations, and a vast ingredient database powered by ML
