IFF SDK benchmarks

This page presents the latest test results of real machine vision cases where IFF SDK is used, such as:

 

RTSP / WebRTC streaming from a camera and host computer to a remote receiving station

Transmitting part with the camera and host computer:
  • XIMEA PCIe machine vision camera
  • NVIDIA Jetson module or desktop computer with NVIDIA GPU
  • Jetson Linux or Windows/Linux for desktop
  • IFF SDK farsight sample application for image acquisition, color pre-processing (black level, auto white balance, demosaicing, color correction, gamma, format conversion), HEVC encoding, and streaming
Remote receiving station:
  • MSI Raider x86 laptop with NVIDIA GeForce RTX 3070 Ti GPU
  • Ethernet / Wi-Fi connection to the host
  • Windows 11 OS
  • IFF SDK imagebroker sample application (receiving images, decoding, rendering to screen)
Resulting parameters when streaming images from the camera to the receiving station screen:

 

  • Camera framerate for 3 ms exposure time and 12-bit raw image mode

  • Image processing time on the host computer for 16-bit processing pipeline / Glass-to-Glass latency in milliseconds

  • NVIDIA Jetson power consumption in watts (if applicable)

Full HD (1920×1080) image stream

Jetson AGX Orin

250 FPS
7 ms / 20 ms
32 W

Jetson Orin NX

160 FPS
10 ms / 25 ms
14 W

Jetson AGX Xavier

180 FPS
10 ms / 25 ms
28 W

Jetson Xavier NX

120 FPS
16 ms / 35 ms
14 W

4K UHD (3840×2160) image stream

x86 host with NVIDIA RTX 2080Ti

200 FPS
10 ms / 30 ms

–

Jetson AGX Orin

120 FPS
20 ms / 50 ms
35 W

Jetson Orin NX

60 FPS
30 ms / 80 ms
16 W

8K UHD (7680×4320) image stream

x86 host with NVIDIA RTX 2080Ti

50 FPS
30 ms / 90 ms

–

Jetson AGX Orin

30 FPS
70 ms / 130 ms
34 W

Click here and contact us to unlock the potential of your system!

Denoise

To remove noise from images, IFF SDK has a denoising module that uses Discrete Wavelet Transform (DWT) and thresholding. There are two denoise options available in the SDK:

  • RGB denoise which is applied after demosaicing (debayering),
  • RAW denoise for Mono and Bayer (raw) images.

See denoise implementation details and parameter descriptions in the IFF SDK technical manual.

Testing results of the denoising module on a system with 4K camera and different NVIDIA GPUs are shown in the table.

Denoise performance for 4K UHD (3840 x 2160) 16-bit images

GeForce RTX 3070 Ti Laptop

  • RAW Denoise: 1.77 ms
  • RGB Denoise: 3.25 ms

GeForce RTX 2080 Ti

  • RAW Denoise: 2.20 ms
  • RGB Denoise: 3.55 ms

Jetson AGX Orin

  • RAW Denoise: 3.72 ms
  • RGB Denoise: 7.19 ms

Jetson AGX Xavier

  • RAW Denoise: 6.76 ms
  • RGB Denoise: 13.45 ms

Image comparison

 

Download the original TIFF images to explore the noise reduction result for a specific case with XIMEA MC124CG-SY color camera as an image source and laptop equipped with RTX 3070 Ti GPU.

Standard image processing pipeline with:

  • 3840 x 2160 ROI on the camera
  • DFPD high-quality demosaic with 11×11 processing window
  • Denoise:
    • without any denoise or
    • with both RAW and RGB denoise options applied
  • Image recording in TIFF format

Without denoise

After RAW and RGB denoise

Comparison of two cropped (900 x 600) images with Before/After image slider:

  • Before – without any denoise
  • After – with both RAW and RGB denoise
without denoisewith RAW and RGB denoise

10 reasons to use

Latest IFF SDK Color Study release

MRTech on Github