Q: What type of licensing options are available for VVAS source code (package)?
VVAS Release has been covered by the following licenses:
Apache License Version 2.0
3-Clause BSD License
GNU GENERAL PUBLIC LICENSE
The MIT License
Q: Which AI models are supported with VVAS?
The following models are supported for this release:
ResNet50
ResNet18
YOLOv5n
YOLOX-M
Q: What is the version of the AMD Vitis™ AI tool used for VVAS?
This VVAS release supports Vitis AI 6.3.
Q: Can the model be changed dynamically?
While a pipeline is running, the model parameters cannot be modified. To change the model’s parameters, stop the running pipeline, update the JSON file, and then re-start the pipeline.
Q: Is receiving RTSP stream supported?
This release does not cover support for RTSP stream.
Q: How do I develop kernel libraries?
Refer to the Acceleration s/w development guide.
Q: Do I need FPGA design experience to develop video analytics applications with VVAS?
No. VVAS SDK ships with most of the building blocks needed for video analytics applications. These building blocks are highly optimized and ready to use. There are several example designs available with this release for video analytics applications. You might directly use these or make modifications as per your needs to build video analytics applications. Refer to the quick-start guide.
Q: Can I generate multiple output resolutions for a single input frame?
Yes. The vvas_xabrscaler plug-in controls the image processing kernel to generate multiple different resolutions for one input frame. This plugin, along with resize, can also do colorspace conversion.
Q: Is audio analytics supported?
No.
Q: Is there support for multi-stage (cascading) inference?
Yes. A cascaded pipeline chains multiple vvas_xinfer elements where the first model (for example, YOLOX-M detector) produces bounding boxes, and a second model (for example, ResNet50 classifier) runs on each detected region. Set inference-level to 2 in the second stage JSON config. Refer to the Cascaded Inference Use Case for a working example.
Q: How to debug VVAS application if there are any issues?
Refer to VVAS Debug Support.
Q: How do I check the throughput of VVAS application/pipeline?
Using GStreamer’s native fps display mechanism.
Q: What is DMA-BUF zero-copy and when should I use it?
DMA-BUF zero-copy allows vvas_xinfer to directly access hardware accelerator buffers without copying data through the CPU. This reduces CPU usage and memory bandwidth. It is enabled by default when using the VART backend. Refer to Buffer Allocation Strategies for details.
Q: How do I verify VVAS plugins are installed correctly?
Use gst-inspect-1.0 to check individual plugins:
gst-inspect-1.0 vvas_xinfer
gst-inspect-1.0 vvas_xabrscaler
If installed correctly, detailed plugin information will be displayed. If you see “No such element or plugin”, check the installation.
Q: How do I measure pipeline latency?
Use GStreamer’s built-in latency tracer:
GST_TRACERS="latency" GST_DEBUG=GST_TRACER:7 gst-launch-1.0 <pipeline>
Refer to Performance Considerations for more details.
Q: What if I cannot find the information that I am looking for?
Contact support at vvas_discuss@amd.com.