AI Application Hub
Explore all application and model setups across industry verticals. Use the sidebar to filter by vertical, model architecture, and target device.
Application Overview
Explore the selected application, model, and deployment fit.
Application Summary
This section summarizes the selected application.
Implementation In Progress
Detailed walkthrough content for this application is being structured. Use the landing cards, filters, and search to compare solution options across verticals.
Defect Detection with Anomalib
Anomaly detection library for manufacturing quality control on AMD hardware
Anomalib Framework
Anomalib is an open-source deep learning library for anomaly detection, developed by Intel and optimized for AMD hardware through ROCm. It provides state-of-the-art anomaly detection algorithms including PatchCore, PaDiM, and STFPM for detecting defects in manufacturing products.
Expected Detection Output
Anomalib provides heatmap-based anomaly localization with pixel-level anomaly scores, enabling precise defect identification on production lines.
Anomaly heatmap visualization placeholder
Target Performance Metrics
Anomalib on AMD embedded hardware delivers real-time anomaly detection with minimal latency.
AMD Ryzen AI Embedded Platform Overview
AMD Embedded Ryzen AI processors deliver a powerful heterogeneous computing platform combining CPU, iGPU, and NPU to enable efficient AI workloads at the edge.
Zen Architecture Cores provide high-performance general-purpose computing for data preprocessing, system control, and post-processing tasks.
- Up to 8 cores / 16 threads
- Optimal for control logic & I/O
- System orchestration
Integrated GPU (RDNA Architecture) delivers parallel compute power ideal for computer vision tasks and deep learning inference.
- RDNA 3 graphics cores
- FP16/FP32 precision
- ROCm stack support
Neural Processing Unit (XDNA Architecture) offers dedicated AI acceleration with extreme power efficiency for INT8 quantized models.
- Up to 50 TOPS INT8
- Ultra-low power (2-5W)
- Optimized for transformers
Heterogeneous Computing Advantage
By leveraging CPU, iGPU, and NPU together, you can optimize your workload placement for the optimal balance of performance, power, and accuracy.
What is AMD NPU Engine?
The AMD NPU (Neural Processing Unit), also known as Ryzen AI, is a dedicated AI acceleration engine integrated into AMD Ryzen processors. It's specifically designed for efficient AI inference workloads, providing hardware-accelerated matrix operations optimized for neural network execution.
Key Features of AMD NPU
- Dedicated AI Hardware: Purpose-built XDNA architecture for AI workloads
- Low Power Consumption: Up to 50% more power efficient than GPU inference
- High Performance: Up to 50 TOPS (Tera Operations Per Second) on Ryzen AI models
- Parallel Processing: Run AI tasks on NPU while GPU handles graphics/compute
Benefits of AMD iGPU
Power Efficiency
Reduce power consumption in 24/7 production line deployments while maintaining high throughput
Resource Optimization
Free up GPU resources for other tasks while NPU handles inference workloads
Consistent Performance
Deterministic latency perfect for real-time quality control applications
Cost Effective
Lower total cost of ownership with reduced power and cooling requirements
Do You Have AMD Embedded Hardware?
This application requires AMD iGPU and/or NPU to run the defect detection models. Select your hardware availability below.
Great! You can proceed with the setup on your local AMD GPU. Make sure you have ROCm drivers installed and configured properly.
• AMD NPU XDNA or XDNA2
cd /usr/local/amd-infra
chmod +x setup.sh && ./setup.sh
No problem! Access a fully configured AMD GPU environment in the cloud with all dependencies pre-installed and ready to run.
Instant access to AMD EPYC CPUs and Radeon GPUs
Defect Detection with RIDAC
Robust Industrial Defect Anomaly Classification on AMD hardware
RIDAC Model
RIDAC (Robust Industrial Defect Anomaly Classification) is a specialized model designed for high-accuracy defect classification in industrial settings. It uses multi-scale feature extraction with attention mechanisms to achieve state-of-the-art accuracy on industrial defect benchmarks like MVTec AD.
Target Performance Metrics
RIDAC is optimized for high-throughput classification on AMD embedded hardware.
Dimension Measurement with PointNet
3D point cloud processing for precision dimensional analysis
PointNet for Dimensional Analysis
PointNet directly consumes 3D point cloud data for segmentation and measurement tasks. Optimized for AMD iGPU, it enables real-time dimensional inspection of manufactured parts, detecting out-of-tolerance measurements at production speed.
Target Performance Metrics
PointNet delivers real-time 3D analysis on AMD embedded hardware.
Dimension Measurement with DeepLabV3
Semantic segmentation for precision edge detection and measurement
DeepLabV3 for Measurement
DeepLabV3 uses atrous spatial pyramid pooling for multi-scale segmentation, enabling precise edge detection required for sub-pixel dimensional measurement. AMD-optimized for real-time part boundary extraction.
Label Inspection with PaddleOCR
Optical character recognition for automated label verification
PaddleOCR for Label Inspection
PaddleOCR is a multilingual OCR toolkit that combines text detection, recognition, and classification. Optimized for AMD hardware, it enables real-time label reading, barcode verification, and date/lot code validation on production lines.
Label Inspection with YOLOv8-Seg
Instance segmentation for label region detection and damage analysis
YOLOv8-Seg for Label Inspection
YOLOv8-Seg combines object detection with instance segmentation to locate labels, detect damage regions, and verify placement accuracy. Runs at real-time speeds on AMD iGPU for inline quality inspection.
Product Sorting with EfficientNet
High-accuracy image classification for automated product sorting
EfficientNet for Product Sorting
EfficientNet uses compound scaling to achieve optimal accuracy-efficiency trade-off for multi-class product classification. Ideal for sorting products by type, grade, or quality level at high conveyor speeds on AMD embedded hardware.
Target Performance Metrics
EfficientNet delivers ultra-fast classification for high-speed sorting lines.
Product Sorting with MobileNetV3
Lightweight classification model optimized for edge deployment
MobileNetV3 for Product Sorting
MobileNetV3 is an ultra-efficient classification model designed for edge devices. Using hardware-aware NAS and squeeze-and-excitation blocks, it achieves excellent accuracy with minimal compute, perfect for NPU deployment on AMD Ryzen AI.