Intel® Geti™ software’s core capabilities empower business and data science users to work together within a single platform. This demo video showcases the capabilities enabling these users with varied expertise utilize the platform to accelerate computer vision workflows.
Transcript:
Edge and AI initiatives are essential for organizational growth and success, yet many face challenges in building effective AI, integrating with existing systems, or scaling globally. The Intel Tiber Edge Platform offers a comprehensive solution to build, deploy, run, secure, and manage Edge and AI solutions at scale.
Creating custom and effective AI models tailored to specific use cases represents a major hurdle in developing these solutions. This is where the Intel Tiber Edge Platform excels, particularly with its Intel Geti functionality. Imagine having the capability to train custom AI models rapidly. What problems could you solve today?
Intel Geti software eases the complexities of developing AI models and enables users to build effective computer vision solutions. The software allows users to start quickly with computer vision tasks, including object detection, semantic and instance segmentation, classification, and anomaly-based tasks.
In complex use cases, a combination of two tasks could solve problems efficiently. Intel Geti software provides a unique ability to combine multiple tasks in the task chain to solve those complex problems and boost collaboration. To create a project, users can simply start with a template, add labels, and upload a few images or short video clips. They can also import projects or bring in an existing dataset and annotations in recognizable formats, such as COCO, YOLO, or VOC.
When beginning a project, the first step is annotating the data. In a typical computer vision pipeline, there are often thousands of objects to annotate. With a wide range of smart annotation features, from bounding boxes, to quick selection, interactive segmentation, and similar object detection, Intel Geti's software expedites and simplifies data tagging with familiar drawing features and AI-assisted labeling.
The software utilizes active learning and guided annotations to intelligently select samples from datasets to reduce sample bias and help models learn faster. Model training starts with as few as 12 annotated images or video frames. Active learning then helps select the most optimal data samples. The trained model makes predictions on those samples and seeks feedback from a human expert. Users can simply accept or edit model predictions to help improve the model's functionality.
The model dashboard offers important statistics to help users make decisions about model readiness. Quantizing models using OpenVINO toolkit is built in, which enables users to optimize inference performance for target hardware in just a few clicks. Users can further evaluate the model performance by running tests on annotated data with either the native framework models or optimized models for the OpenVINO toolkit.
The software allows users to discover and optimize the best set of learning parameters for their AI models. The software supports multiple model architectures to enable users to customize models based on their application requirements without needing to start from scratch. Users can also optimize the training parameters, either manually or by using the automated optimization feature.
Furthermore, the available software development kit makes REST APIs accessible for developers to connect their pipeline to the software, upload new data, design training experiments, and export the newly-trained model when available to their deployment pipeline. Once the AI model is ready, users can download it in a few clicks from the interface or via the SDK. The sample code with the downloaded model helps users run the optimized model within just a few steps and utilizes OpenVINO's APIs, making the deployment agnostic to the model architecture or computer vision task.
Models built with Intel Geti software can be seamlessly integrated into applications and solutions with pre-built micro-services, such as model registry, and can be easily deployed and scaled across distributed Edge and cloud locations using the Intel Tiber Edge Platform. The Intel Tiber Edge Platform and Intel Geti software represent a revolutionary new approach to computer vision model development and implementation.
Learn more about how you can accelerate the speed of AI development and deployment today.