Savant

• High-performance real-time computer vision and video analytics pipelines
• NVIDIA DeepStream integration for accelerated video processing and inference
• Support for NVIDIA data center GPUs and Jetson edge devices
• Python-based development with YAML pipeline configuration
• Integration with OpenCV CUDA, PyTorch, CuPy, and CV-CUDA
• Object detection, classification, segmentation, tracking, and custom processing
• Support for RTSP, USB, CSI-2, GigE Vision, and video file sources
• Docker-based deployment for reproducible computer vision applications
• Built-in adapters for video, image, metadata, and streaming workflows
• Scalable edge, data center, and hybrid computer vision deployments

Savant is an open-source framework designed for developers building real-time computer vision and video analytics applications on NVIDIA hardware. Rather than functioning as a single computer vision library, Savant provides an application architecture for creating complete processing pipelines that combine video ingestion, AI inference, tracking, metadata processing, streaming, and output.

The framework is built on NVIDIA DeepStream and is designed to simplify the development of high-performance streaming AI applications. Developers can define pipeline structures using YAML configuration while implementing custom processing logic with Python. This approach allows teams to concentrate on application-specific computer vision functionality without having to build the underlying pipeline architecture from scratch.

Savant supports NVIDIA data center GPUs as well as NVIDIA Jetson edge devices. It can be used for applications involving object detection, classification, segmentation, tracking, video processing, and other computer vision workloads. The framework can integrate technologies including OpenCV CUDA, PyTorch, CuPy, and CV-CUDA, providing flexibility when creating customized AI processing pipelines.

The platform also supports multiple video and image input methods, including RTSP streams, USB cameras, CSI-2 cameras, GigE Vision devices, and video files. Its Docker-based runtime helps package and deploy modules consistently across supported environments.

Savant is intended primarily for computer vision engineers, AI developers, researchers, and technical teams building production-oriented video analytics systems. Example applications include people detection, vehicle analysis, traffic monitoring, industrial monitoring, and other real-time visual processing scenarios.

Savant is free to use as open-source software under the Apache 2.0 license. The project also offers custom services for organizations requiring specialized computer vision development or implementation support.

Frequently Asked Questions

What is Savant?

Savant is an open-source framework for building real-time computer vision and video analytics applications on NVIDIA hardware.

Is Savant an AI model?

No, Savant is a computer vision pipeline framework rather than a standalone AI model, allowing developers to combine models and processing components into complete applications.

Is Savant free?

Yes, Savant is free and open-source software distributed under the Apache 2.0 license.

What hardware does Savant support?

Savant supports NVIDIA data center GPUs and NVIDIA Jetson edge devices for accelerated computer vision workloads.

What programming languages does Savant use?

Savant uses Python for application-specific processing and also incorporates C++ and Rust components within its underlying framework.

Can Savant process live video streams?

Yes, Savant can process live streams and supports sources such as RTSP, USB, CSI-2, and GigE Vision cameras.

Does Savant support custom AI models?

Yes, Savant provides a pipeline architecture that allows developers to integrate custom models and processing components into computer vision applications.

Who is Savant designed for?

Savant is designed for computer vision engineers, AI developers, researchers, and technical teams developing real-time video analytics applications.

Share Your Experience: