Adlink spins edge vision SDK for Nvidia and Intel

Adlink spins edge vision SDK for Nvidia and Intel

Sep 8, 2021 — by Eric Brown

— 138 views

Adlink announced an Ubuntu-ready “Edge Vision Analytics” (EVA) SDK with a GUI-based IDE and open source plug-ins for Nvidia and Intel systems. EVA is initially available with its Neon-2000-JNX camera and EOS-i6000-P computer.

In May, when Adlink announced its Jetson Xavier NX based NEON-2000-JNX embedded camera, the company mentioned support for an Edge Vision Analytics (EVA) SDK. We assumed this was just another name for an Adlink spin of Nvidia’s Ubuntu 18.04 based L4T (Linux for Tegra) stack with JetPack AI tools. Adlink has now formally announced EVA, and it turns out it is more than an Nvidia wrapper, as it also supports Intel CPU, iGPU, and VPU AI acceleration, as well as ONNX runtime, TensorRT, and Intel’s OpenVINO technologies.

EVA SDK’s AI Pipeline Studio
(click image to enlarge)

The EVA SDK lets users set up a demo in two days, a PoC (Point-of-Concept) in two weeks, and deployable hardware and software in two months, claims Adlink. The SDK is built around a GUI-based, drag-and-drop AI Pipeline Studio IDE aimed at bringing AI novices quickly up to speed.

The AI Pipeline Studio provides for easy switching of image sources, image processing adjustment, and AI models. It enables a review function for each step, including image capturing, image pre-processing, and AI inference for easy verification and debugging, says Adlink.

The platform supports both Ubuntu 18.04 and Windows 10 and provides open source, field-ready plugins for hardware codec, image processing, and edge communication. There are optimized models for object classification, detection, segmentation, and human posture detection. The platform can accommodate hybrid AI inference engines simultaneously in a pipeline.

The EVA SDK supports 10+ types of USB Vision, GiGE Vision, and MIPI-CSI-connected cameras. Specific brands include Basler, FLIR, and Hikrobot. There is also support for V4L2 and RTSP, as well as connectors for AWS, DDS data river, ROS2, and Microsoft Azure.

NEON-2000-JNX (left) and EOS-i6000-P
(click images to enlarge)

EVA is initially available for the NEON-2000-JNX and the earlier and very similar NEON-2000-JT2, which also incorporates the Xavier NX. The Neon-2000-JNX is a C-mount camera with GbE, USB Type-C with DP, and support for 6x sensors up to an 8MP Sony IMX334.

EVA also supports Adlink’s Intel 9th Gen Coffee Lake based EOS-i6000-P computer, as well as an EOS-JNX-I/EOS-JNX-G product line that has yet to be announced. The EOS-i6000-P is a feature-rich system aimed at GiGE Vision applications with multiple full-sized slots for Nvidia Quadro GPU cards. Similar Adlink Coffee Lake systems include the Matrix MVP, MXC, and MXE computers and DLAP-8000 robotics controller.

“EVA SDK is a one-API framework that lets users build edge AI vision applications with low coding efforts,” stated Chia-Wei Yang, Head of Edge Vision Business Center, Adlink. “Even without prior, or deep AI knowledge or expertise, users can develop an edge AI POC with the ready-to-use plugins and reference codes using the drag-and-drop GUI, which ultimately speeds up deployment and time to market.”

Further information

The EVA SDK is available now for Adlink’s NEON-2000-JNX, NEON-2000-JT2, EOS-i6000-P, and EOS-JNX-I/EOS-JNX-G, with pricing undisclosed. More information may be found in Adlink’s announcement and product page, as well as the GitHub page, which offers downloads of open source “Showcase” plug-ins.