Image delivery
Image processing and delivery is the most common task in machine vision systems, and our team focuses on it. Below are some examples of how we solve these kinds of tasks.
Image delivery solutions from MRTech are designed to receive images from one or more cameras, pre-process the images on various hardware platforms using the required processing pipeline, and then deliver them to their destination.
On our GitHub account, you can find such simple applications for the C/C++/Python APIs from MRTech IFF SDK.
How IFF SDK works with Blackwell GPU
It comes as no surprise that implementing the IFF SDK application on a system with the most powerful GPU delivers the highest performance. We measured the performance of Linux working station equipped with NVIDIA RTX PRO 6000 Blackwell Workstation Edition GPU and with four 65MP XIMEA cameras connected to the computer.
A standard image processing pipeline was used:
- 8K image acquisition from one, two, three and four cameras
- Image pre-processing including demosaic and color correction
- H.265 encoding and RTSP streaming
So, this configuration is capable of processing the following image streams:
- 2x 8K UHD (7680×4320) 10-bit images @ 110 FPS or 7.3 Gigapixels/s (limited by maximum camera frame rate)
- 3x 8K UHD (7680×4320) 10-bit images @ 90 FPS or 9.0 Gigapixels/s
- 4x 8K UHD (7680×4320) 10-bit images @ 65 FPS or 8.6 Gigapixels/s
See the full test results via the link ->
IFF SDK extra performance on NVIDIA Jetson
System configuration:
- Modern NVIDIA Jetson modules: Orin NX / AGX Orin / AGX Thor
- High-performance PCIe XIMEA cameras
- IFF SDK `farsight` sample application with the processing pipeline:
- Image acquisition from the cameras
- Image pre-processing including demosaic and color correction
- H.265 encoding and RTSP streaming
Image delivery for AI works on a wide range of CPU / GPU systems
Digital zoom for an object tracking
Digital zoom is a method of enlarging a section of an image or video by cropping the original high-resolution image. This allows, for example, a remote drone operator to examine a specific area of the captured image in great detail, by reducing the field of view and increasing the resolution. It is important, however, to have a high-resolution input image and a high-performance system.
AI accelerator on NXP Semiconductors platform
This system run an anonymization application with the hardware acceleration of neural network algorithms, including person detection, pose estimation, hand bone position estimation, face detection, and face expression classification. It is based on I-Pi SMARC development kit powered by NXP i.MX 8M Plus module with a miniature MIPI camera from Vision Components connected.
AI with Qualcomm
This setup with Qualcomm RB3 Gen 2 Core Kit and XIMEA machine vision camera highlights the following features:
- Common IFF SDK use case – image delivery to AI application based on artificial neural networks or other machine learning and intelligence technologies.
- Vulkan IFF SDK component, bringing support for image processing on a wide range of GPUs.
- Integration of IFF SDK and GStreamer pipelines, which allows to re-use parts of existing GStreamer-based solutions, while utilizing IFF SDK capabilities to interface with machine vision cameras and to quickly pre-process acquired images.
- IFF SDK support for various hardware platforms (Intel and ARM architectures) and operating systems (Linux and Windows), including Yocto-based Linux ARM64 distributions.









