Overview

FFNet Realtime Semantic Segmentation on Qualcomm® Hexagon™

Short summary: FFNet delivers real-time semantic segmentation on Qualcomm Hexagon platforms, using ONNX Runtime QNN and QAIRT for NPU-accelerated urban scene understanding with enhanced visual legend and boundary rendering.

About Advantech Container Catalog (ACC)

Advantech Container Catalog is a comprehensive collection of ready-to-use, containerized software packages designed to accelerate the development and deployment of Edge AI applications. By offering pre-integrated solutions optimized for embedded hardware, it simplifies the challenges often faced with software and hardware compatibility, especially in GPU/NPU-accelerated environments.

Feature / Benefit	Description
Accelerated Edge AI Development	Ready-to-use containerized solutions for faster prototyping and deployment
Hardware Compatible	Reduces hardware and package incompatibility issues
GPU/NPU Access Ready	Supports passthrough for efficient hardware acceleration
Model Conversion & Optimization	Built-in model conversion and quantization recommendations
Optimized for CV & LLM Applications	Optimized stacks for vision and language workloads

Container Overview

This container validates Qualcomm NPU-enabled ONNX Runtime on Hexagon platforms and performs real-time semantic segmentation using FFNet. It provides a containerized environment with custom onnxruntime-qnn, QAIRT, and LiteRT support for high-throughput inference on Advantech AOM-2721 and AIR-055.

Demo

Use Case

Real-time semantic segmentation for urban scene understanding
Embedded traffic and video analytics on Qualcomm Hexagon devices
NPU-accelerated batch video processing with enhanced visualization
Edge AI deployment on Advantech AOM-2721 and AIR-055 platforms

Key Features

FFNet realtime semantic segmentation using ONNX Runtime QNN
Qualcomm Hexagon NPU acceleration with W8A8 quantized models
Simple container-based deployment and CLI video inference workflow

Host Device Prerequisites

Item	Specification
Compatible Hardware	Advantech AOM-2721 or AIR-055 with Qualcomm QCS6490 / IQ-9075 and Hexagon™ DSP
Platform Version	Ubuntu 22.04 guest OS
Host OS	Linux on Qualcomm development board
Required Packages	Git, Docker Engine, container runtime, onnxruntime-qnn environment
Software Installation Guide	README.md in repository

Required Software Packages on Host Device

Component	Version	Description
Ubuntu	22.04	Guest OS
Python	3.10	Runtime environment
ONNX Runtime (QNN EP)	1.24.1	Custom build with QNN Execution Provider
QAIRT (QNN SDK)	2.43.0	Qualcomm AI runtime backend
LiteRT	2.1.4	QNN TFLite Delegate support for GPU/NPU acceleration

Container Environment Overview

Software Components in the Image

Component	Version	Description
Ubuntu	22.04	Base guest OS
Python	3.10	Runtime environment
ONNX Runtime (QNN EP)	1.24.1	Inference runtime for ONNX models
QAIRT	2.43.0	Qualcomm AI runtime backend
LiteRT	2.1.4	GPU/NPU runtime support for QNN delegate

Container Quick Start Guide

For installation, setup, build scripts, and detailed usage instructions, please refer to the FFNet Realtime Semantic Segmentation on Qualcomm® Hexagon™ in the repository.

Supported AI Capabilities

Vision Models

Model Family	Versions	Notes
FFNet	W8A8 quantized	Real-time semantic segmentation on Qualcomm NPU
Semantic Segmentation	ONNX	Cityscapes-style 19-category output
Boundary Enhancement	Built-in	High-contrast contour rendering

Supported AI Model Formats

Format	Support Level	Notes
ONNX	Full	Native ONNX Runtime QNN support
TensorRT™	N/A	Not used in this container
PyTorch (JIT)	N/A	Not used in this container
TensorFlow SavedModel	N/A	Not used in this container

Hardware Acceleration Support

Accelerator	Support Level	Compatible Libraries	Notes
Qualcomm Hexagon NPU	Full	ONNX Runtime QNN, QAIRT	Best performance with W8A8 models
CPU	Full	ONNX Runtime	Fallback mode
Adreno GPU	Partial	LiteRT / QNN delegate	Indirect support via Qualcomm runtime

Troubleshooting & Notes

Common issues:
- Custom onnxruntime-qnn build only works inside this container environment.
- Ensure the target board is AOM-2721 or AIR-055 with the correct Hexagon runtime installed.
- Use chmod +x on scripts and verify the container launched successfully.
Known limitations:
- Tested on AOM-2721 and AIR-055 only.
- NPU acceleration results are based on W8A8 quantized models.
- Not a generic x86 container image.