Previous Releases - Previous Releases - 6.3 English - XD056

Vitis Video Analytics SDK (VVAS) 6.3 User Guide (XD056)

Document_ID: XD056
Release_Date: 2026-09-24
Version: 6.3 English

Vitis Video Analytics SDK Overview
Vitis Video Analytics SDK Overview
Vitis Video Analytics SDK Overview
VVAS Graph Architecture
Why VVAS?
VVAS Components
Custom GStreamer Plug-ins
Infrastructure GStreamer Plug-ins
Acceleration S/W Libs
Acceleration H/W (Kernels/IPs)
Reference Applications
Release Notes
Vitis Video Analytics SDK Release V6.3
Vitis Video Analytics SDK Release V6.3
Release Summary
New Features
Updates
Known Limitations
Previous Releases
Vitis Video Analytics SDK Release V6.2
Release Summary
Features
Known Limitations
Quick Start Guide
Quick Start Guide for VVAS
VEK385 Platform Variants
Prerequisites and System Image Flashing Steps
Running the Solution from the Pre-Built VVAS
Quick Start Examples
Extended Examples
Single-Instance Inference
Multi-Instance Inference
Compiling VVAS from the Source Code
Set Up Cross Compilation Environment and Compilation
Build VVAS
VVAS Use Cases
VVAS Use Cases
Multiple Inference Instance Deployment
Using Single Process
Using Multiple Processes
Spatial Sharing Use Case
Temporal Sharing (Multi-Tenant) Use Case
Zero Copy Use Case
Cascaded (Multi-Level) Inference Use Case
Four-Camera MIPI Streaming Use Case
Example Scripts
Base Infrastructure
GStreamer Plug-ins
Custom Plug-ins
vvas_xmetaaffixer
Input and Output
Control Parameters and Plug-in Properties
Pad Naming Syntax
Example Pipelines
vvas_xabrscaler
Prerequisite
Input and Output
Control Parameters and Plug-in Properties
Example Pipelines
One input One Output
One Input Multiple Output
Crop with Multiple Outputs
vvas_xinfer
Key Features
Key Pre-Processing Features
Post Processing
Key Features of GstInferenceMeta
Wide Input Format Support
Tensor Video Format Names
Control Parameters
Signals
JSON File Format to Be Provided as Input to the config-file Property
config-file json members
config-file infer-config JSON Members
config-file preprocess-config JSON Members
config-file postprocess-config JSON Members
Example pipelines
Single stage inference pipeline
Multi-level cascaded inference pipeline
Metadata tree (single frame with two detections and two classification results)
Yolo + Resnet50 + pipeline
Yolo + Resnet50 + Resnet50 (L2) pipeline
Advanced Features
Batch Processing Support
VART Post-Processing Support
Mixed-Precision Models
Quantization and Dequantization
Profiler Support
Profiler Output JSON Structure
CPU Subgraph Support
Debugging Tips
vvas_xoverlay
Supported Features
Input and Output
Control Parameters and Plug-in Properties
Example Pipelines
vvas_xmetaconvert
Input and Output
Control Parameters and Plug-in Properties
JSON File Format to Be Provided as Input to the config-location Property
Object Detection and Classification
Segmentation
Example Pipelines
vvas_xmulticrop
Static Cropping
Dynamic Cropping
Prerequisite
Input and Output
Control Parameters and Plug-in Properties
Example Pipelines
vvas_xtilecompositor
Supported Features
Prerequisites
Input and Output
Control Parameters and Plug-in Properties
Example Pipelines
Four-camera V4L2 composition
vvas_xfunnel
Control Parameters and Plug-in Properties
Example Pipelines
vvas_xdefunnel
Control Parameters and Plug-in Signals
Example Pipelines
Infrastructure Plug-ins and Acceleration Software Libraries
vvas_xfilter
JSON File Schema
Development Guide
Acceleration Software Library Development Guide
Types of Kernels
Hard Kernel
Software Kernel
Kernel Management Modes
Interfaces for Acceleration Software Libraries
Core API
xlnx_kernel_init
xlnx_kernel_start
xlnx_kernel_done
xlnx_kernel_deinit
Utility APIs
Memory Management APIs
Read/Write Register APIs
Execution APIs
vvas_kernel_start
VVAS Data Structures
VVASKernel
VVASVideoFormat
VVASFrame
Other VVAS Declarations
Acceleration Software Library for Hard Kernels
Memory Allocation
Starting the Kernel
XRT Managed Kernel
User Managed Mode
Checking the Kernel Task Status
XRT Managed Kernel
User Managed Mode
Acceleration Software Library for the Software Kernel
Capability Negotiation Support in the Acceleration Software Library
Example using xstub (Development of an Acceleration Software Library)
Adding New Postprocessing Implementation to VvasPostprocess
Sample Postprocess Library Configuration Reference
Classification: ResNet50 / ResNet18
Detection: YOLO Family
YOLO Output Tensor Layout
YOLO Presets
Practical Guidance
Adding a Custom Postprocess Library
VVAS XInfer Buffer Allocation Strategies
Allocator Negotiation
Negotiation Inputs
Sink Caps
Allocator Proposal
Scenario Matrix
PPE Enabled
System-Memory Input
VVAS-Memory Input
Generic DMA-BUF Input
PPE Disabled
VART HW Backend
ONNX Runtime Backend
Synchronization
Operational Notes
Validation Checklist
Implementation Points
Custom AI Tensor Formats
Naming Convention
GStreamer Names and VVAS Names
Formats Used by vvas_xinfer
Formats Used by vvas_xabrscaler
Supported Format Lists Are Configuration Dependent
VVAS to GStreamer format strings
VVAS Pipeline Performance Considerations
Measure the Complete Pipeline
Measure Pipeline Latency
Check the vvas_xinfer Profiler
Understand Inference Backend Runtimes
Avoid File I/O Bottlenecks During Measurement
Understand Batch and Postprocess Cost
Preserve Zero-Copy Paths
Use Queues and Sinks Carefully
Summary Checklist
VVAS Debugging Support
Enable GStreamer Debug Logs
Run the GStreamer Pipeline with Debug Logs
Analyze the Debug Logs
VVAS Core Debug Logs
Standalone VVAS Core Logging
VVAS Metadata
VVAS Meta Data Structures
VVAS Inference Metadata
GstInferenceMeta
GstInferencePrediction
VvasInferPrediction
VvasInferResultType
VvasInferResult
VvasBoundingBox
VvasInferDetection
VvasInferClassification
GstInferenceMetaData APIs
VVAS Overlay Metadata
GstVvasOverlayMeta
FAQ
Frequently Asked Questions
Troubleshooting Guide
1. VVAS GStreamer Commands Failing - Verify Board and Environment Setup
2. Installation or gst command execution errors
3. Error with xinfer command execution
4. xinfer onnxruntime switching to cpu mode when json is configured to use vitisai
5. Couldn’t create VVAS context / failed to do initialization
6. kernel start failed / failed to process frame in PPE
7. Model compilation is not supported in a deployment only installation
8. vvas_postprocess_create failed
9. infer config file parse error
10. Model resources missing - Neither partition directory nor RAI file found
11. preprocess-config errors
12. Pipeline runs but classification results are incorrect
13. Processing a specific number of frames
14. Missing metadata/overlay on some frames with vvas_xmetaaffixer (tee back-pressure)
Additional Information
Previous Releases
Previous Releases
Vitis Video Analytics SDK Release V5.1
Release Summary
Features
Known Limitations

Back to home page