Edge AI Performance on NVIDIA Jetson: Mastering Orin Nano and TensorRT for Real-Time Computer Vision and Robotics Projects (Edge AI Mastery: Building Intelligent IoT and TinyML Applications) - Softcover

Inhaltsangabe

Unlock the Full Raw Power of the NVIDIA Jetson Orin Nano Super.
The world of AI is moving away from the cloud. The real revolution is happening at the Edge. But having the hardware isn't enough—you need to know how to squeeze every single TOPS of performance out of it.
Whether you are building autonomous drones, industrial robots, or local Generative AI systems, Edge AI Performance on NVIDIA Jetson is the definitive engineering blueprint for high-speed, low-latency deployment.

Why this book is the "Industry Bible" for Jetson Developers:
While others struggle with thermal throttling and "out of memory" errors, you will learn the silicon-level secrets to professional optimization. This isn't a beginner’s "unboxing" guide—it is a deep dive into the JetPack 6.x ecosystem designed for developers who need their models to run 7x faster.

Inside, you will master:

• The TensorRT Mastery: Convert PyTorch and ONNX models into high-performance engines using INT8 Quantization and Entropy Calibration.
• Generative AI at the Edge: Deploy Llama 3 and Mistral locally. Learn to fit 7B parameter models into 8GB RAM using 4-bit AWQ.
• Vision AI Domination: Benchmarks and deployment strategies for YOLO11, real-time segmentation, and pose estimation.
• Zero-Copy Architectures: Eliminate the "Memory Wall" by mastering Unified Memory and pinned memory buffers.
• Isaac ROS & Robotics: Build hardware-accelerated vSLAM and navigation stacks that offload heavy compute to the DLA and PVA.
• Industrial-Grade Security: Protect your IP with Secure Boot, Disk Encryption (LUKS), and Trusted Execution Environments (TEE).

1. Smart City: Multi-camera traffic analytics at 30+ FPS.
2. Agri-Tech: Autonomous weed detection and precise spraying robots.
3. Med-Tech: Real-time surgical tool tracking with sub-10ms latency.

Die Inhaltsangabe kann sich auf eine andere Ausgabe dieses Titels beziehen.