NVIDIA has announced the launch of JetPack 2.3, the latest software suite for developer tools and libraries for its Jetson TX1 embedded system developer kit, which is capable of running complex, deep neural networks.
Available now for free download, JetPack 2.3 is an all-in-one package that bundles and installs all system software, tools, optimized libraries, and APIs, along with various examples. The release supports both the latest Jetson TX1 developer kit, running 64-bit Ubunti 16.04, as well as the Jetson TK1 developer kit, running 32-bit Ubuntu 14.04.
Key features in the release include:
TensorRT:Formerly known as GIE, TensorRT is a deep learning inference engine that maximizes runtime performance for applications such as image classification, segmentation, and object detection. This feature enables developers to deploy real-time neural networks powered on Jetson and offers double the performance for deep learning over previous implementations of cuDNN.
cuDNN 5.1: A CUDA-accelerated library for deep learning that provides implementations for standard routines such as convolutions, activation functions, and tensor transformations. Support for advanced networks models such as LSTM and RNN have also been included in this release.
Multimedia API: A package of low-level APIs, including a camera API that offers per frame control over camera parameters and EGL stream outputs that allow interoperation with GStreamer and V4L2 pipelines as well as access to connect camera sensors over MIPI CSI. Additionally, a V4L2 API offers video decode, encode, format conversion, and scaling functionality and opens up low-level access for features such as bit rate control, quality presets, low latency encode, temporal tradeoff, and motion vector maps.
CUDA 8: JetPack 2.3 also includes updated host compiler support of GCC 5.x and the NVCC CUDA compiler has been optimized for up to 2x faster compilation. CUDA 8 also includes nvGRAPH, an accelerated library for graph analytics, as well as new APIs for half-precision floating point computation in CUDA kernels and cuBLAS and cuFFT libraries.
NVIDIA also announced that it is working withLeopard Imaging Inc., a Jetson Preferred Partner specializing in the creation of camera solutions including stereo depth mapping for embedded machine vision applications, cameras for intelligent machines, and consumer products. Developers can work with Leopard Imaging to integrate multiple RAW image sensors while using NVIDIA’s internal on-chip ISPs or external ISPs via CSI or USB interfaces, in addition to the ISP bypass imaging modes available today for YUV sensors.
Learn more: search the Vision Systems Design Buyer's Guide for companies, new products, press releases, and videos