Latest Release: Savant 0.2.7 Simplifies Use of Nvidia Deepstream for Machine Learning
A new issue of the Python-framework Savant has been published, bringing version 0.2.7 to the table. This release aims to simplify the use of Nvidia Deepstream for solving machine learning-related problems. With Savant, users can now focus on building optimized output conveyors using declarative syntax (YAML) and Python functions, while the framework takes care of all the complex work with Gstreamer or FFMPEG.
Savant also allows users to create conveyors (Pipeline) that work equally well on accelerators in the data center (like NVIDIA TURING, AMPERE, and HOPPER) as well as on EDGE devices such as Jetson NX, AGX XAVIER, Orin NX, AGX Orin, and the new Nano. Additionally, Savant enables users to easily process multiple video streams simultaneously and quickly create video analytics conveyors that are ready for deployment in working applications by using NVIDIA Tensort.
The project code for Savant is open source and distributed under the Apache 2.0 license. You can find more information about the framework and access the code on GitHub.
Regarding the release, Savant 0.2.7 is the last release in the 0.2.x branch that includes functional changes. Future releases in the 0.2.x branch will only address bug fixes. New features will be developed in the 0.3.x branch, which is based on Deepstream 6.4. It’s important to note that the 0.3.x branch will not support the Jetson Xavier family, as NVIDIA has ceased support for them in DS 6.4.
Main Innovations in Savant 0.2.7
- New usage examples:
- An example of working with a detection model based on the transformer RT-Detr
- Cuda post Processing using Cupy for Yolov8-Eg
- An example of the Pytorch Cuda integration into Savant Paydaine
- Demonstration of work with oriented objects