Compatibility

This version of DetectorPack is compatible with Axxon One 2.0 and later versions.

Optimization and improvements

Added support for the Nvidia GPU cards of the Hopper and Blackwell microarchitecture in the Windows OS for the following detectors:

• Neural tracker
• Neural counter
• Equipment detector
• Human pose detector
• Fire detector
• Smoke detector
• Neural classifier
• Barcode detector
• Water level detector
• Stopped object detector
• Privacy masking
• Person-based privacy masking

Update

Recommendations

1. For Linux OS: if before this update, you installed only one detector package that included the whole analytics (not divided into plugins), then before updating, you must remove the installed package without using purge (complete package deletion) by running the following command:
   

   sudo dpkg --remove --force-depends axxon-detector-pack

   and then install the necessary plugins.
2. For Linux OS: you must first install the RR SDK Pack before you install packages for the following detectors (see Installing DetectorPack add-ons):
   • License plate recognition RR (RR License plate recognition add-on);
   • License plate recognition RR—Search in archive (RR License plate recognition add-on);
   • License plate recognition RR—Parking (RR License plate recognition add-on);
   • Vehicle recognition RR (RR Vehicle Recognition add-on).

Limitations

1. When you update Face Recognition Pack add-ons (see Installing DetectorPack add-ons) from version 3.13.2 or earlier, the results of the previously recognized persons will become unavailable. If you want to keep the previously recognized persons, don't update the analytics package.
2. Starting with the 2.0.16 version, Axxon One supports the Nvidia GPU cards of the Blackwell microarchitecture.
3. The following microarchitectures of the Nvidia GPU cards are no longer supported (for more details, see here):
   • Kepler (late),
   • Maxwell,
   • Pascal,
   • Volta.
4. The mechanisms of starting neural models are changed so the part of specialized models (trained for an object, see Data collection requirements for neural network training) can operate improperly. If there are any difficulties, contact the technical support.