Raven AI Engine is a flexible AI framework with turnkey solutions, specially designed for the real-time execution of AI algorithms in multimedia applications. It is the result of combining decades of experience in video and multimedia with our expertise in artificial intelligence.
With our in-house technology, we ensure exclusive solutions and customization
We tailor the engine to client needs, optimizing performance, accuracy, and speed based on project and hardware requirements.
Our in-house engine allows us to stay ahead of AI and video processing trends, delivering cutting-edge solutions.
As the engine's developers, we offer specialized support and rapid, customized maintenance when needed.
Developed in-house, Raven AI combines advanced machine learning algorithms with a versatile, multi-platform framework, offering exceptional performance and adaptability. With full control over the technology stack, we can customize solutions for specialized hardware, operating systems, or unique environments.
Our engine is much more than an AI runtime toolkit; it's a complete SDK offering advanced graphical pre- and post-processing capabilities, GStreamer integration, and multimodal AI inference. It utilizes hardware acceleration to maximize performance and ensure real-time efficiency across demanding tasks.
Raven AI integrates with multiple APIs and programming languages, easily incorporating it into existing workflows and systems. This flexibility allows businesses to adopt this technology without overhauling their current infrastructure.
With bindings available for C++, Python, and C# and integration with GStreamer, clients can easily embed Raven AI Engine into their existing codebases, eliminating the need for significant application changes. This ensures a smooth and efficient integration process, minimizing disruption while maximizing the technology's value.
With Raven, the client code is written once and compiled without changes across multiple hardware environments and operating systems. It automatically selects the best available hardware on the machine, whether it’s AMD, Intel, or NVIDIA GPUs, ensuring an optimal workflow pipeline. Additionally, our engine supports video from multiple graphic APIs, such as DirectX12 and Vulkan, allowing for more efficient operation across a wide range of applications and systems.
// Inputs setting
fsai::Image image(file);
auto engine = fsai::EngineFactory::createByType("dml");
auto texture = engine->createTexture(image);
// Specific API setting
fsai::sports::strongsort::ModelDetectionFastGeneric model{“models”};
fsai::sports::strongsort::DetectorFastGeneric detector{modelGeneric, engine};
// Inference
auto detections = detector.push(texture);
// Get bounding boxes
for (auto& detection : detection){
const auto& coords = detection.getBoundingBox() ; }
Stay updated with news, our latest projects, insights, articles, and open-source contributions from the Fluendo team.
In this post, we explore the transition from traditional manual sports video analysis to cutting-edge autonomous scene calibration.
Explore in this entry how sports analysis has evolved with computer vision, from manual scrutiny to AI integration.
Dive into this article, where we showcase significant improvements in detection accuracy and tracking performance, and learn about our ongoing commitment to innovation in multimedia edge AI.
Discover how Fluendo's Innovation Days have unleashed the power of post-training optimization for AI at the edge in this blog.