Emergent Adds New Plug-ins With GPUDirect for eCapture Pro
PORT COQUITLAM, British Columbia — Jan. 30, 2024 — Emergent Vision Technologies, a pioneer in high-speed GigE Vision cameras and zero-copy and zero-loss vision technologies, introduces new plug-ins that leverage GPUDirect technology, which supports Emergent’s renowned zero-copy, zero-loss imaging approach. The new built-in plug-ins include polarization, H.265/RTMP (real-time messaging protocol), pattern matching, and inference.
Supported by eCapture Pro software, GPUDirect technology enables the transfer of images from the camera directly to the GPU, which bypasses system memory and the CPU, delivering zero-copy and zero-loss imaging capabilities. In addition, Emergent offers optimal flexibility in camera stream routing to any GPU in a system, which simplifies processing distribution. Supported GPUs include NVIDIA RTX A6000, RTX A5000, RTX A4000, RTX 6000 ADA, Jetson Orin, and Jetson Xavier.
“With the tremendous amounts of data that Emergent’s GigE Vision cameras capture, users need a means for processing that data with top performance,” said John Ilett, president and CTO at Emergent Vision Technologies. “Our GPU plug-in technology is supported within multi-camera and multi-server systems for maximum performance and ease of scalability.”
End users can also create custom plug-ins in eCapture Pro that leverage Emergent’s zero copy, zero loss, GPUDirect approach while only writing plug-in code. In the link below, the demonstration shows the ease with which a plug-in can be deployed. First, CUDA-based code is written so the plug-in can run on an NVIDIA GPU. The code is then compiled to create the plug-in DLL and loaded into the software using a simple drag and drop interface. Lastly, the plug-in is instantiated on the selected GPU, allowing the end user to run the plug-in within eCapture Pro software.
In eCapture Pro software, end users can leverage the new built-in plug-ins, such as the polarization tool. This tool lets users review the benefits of the characteristic outputs of a standard polarized processing pipeline, such as degree of polarization or direction or angle of polarization. End users can even remove the polarized light or output one of the four orientation options: 0, 45, 90, or 135 degrees.
Deep Learning Inference Plug-in
With the inference plug-in, users can easily add and test their own trained deep learning inference model to perform detection and classification of objects. After training a model in a framework such as PyTorch or TensorFlow, users can add the model to the plug-in, instantiate it on the desired GPU, connect it to the camera, and click “run” to begin, allowing the software to detect and classify objects even while in motion.
Pattern Matching Plug-In
To use the pattern matching plug-in within eCapture Pro, the user first creates a pattern template, loads the plug-in, and then instantiates it on the desired GPU. Next the user specifies the path to the pattern template, connects the camera to the GPU, and runs the plug-in to begin matching patterns, including those on moving objects.
Emergent’s eCapture Pro software also offers a plug-in for the H.265 video codec, which is the successor to H.264, delivering up to 50% better video compression while maintaining the same level of video quality, making it an ideal option for high-speed, high-resolution video. In addition, the plug-in supports RTMP streaming, which enables users to stream high-speed video to YouTube and other live streaming platforms. In high camera count systems, this plug-in can help users handle vast amounts of data. In a 48-camera system leveraging a midrange server, a 48-port switch, and two GPUs, for example, all images can be compressed and stored on a single local M.2 drive while one camera from the system streams to an RTMP client such as YouTube.