Course Objective & Scope
A vision model that runs accurately on a data-centre GPU often has to run directly on an edge device in the field because of bandwidth costs, intermittent connectivity, strict latency targets or data privacy. Starting from exporting PyTorch models to ONNX, this programme covers compiling them for target hardware using NVIDIA TensorRT, Intel OpenVINO and TFLite/ExecuTorch.
FP16 and INT8 quantization calibration, hardware-accelerated camera ingestion (CSI/RTSP + GStreamer), memory and thermal bottleneck profiling, and remote fleet monitoring and updates are covered hands-on. Participants leave with a complete edge deployment plan and an optimised inference package for their own models.
Who Should Attend?
- AI engineers deploying vision models onto embedded systems and field appliances
- Embedded software developers working with NVIDIA Jetson, industrial PCs or mobile/IoT hardware
- MLOps and platform engineers responsible for field operations of camera-based products
- Prerequisite: Python, PyTorch and basic Linux/container usage
Key Takeaways
- Convert PyTorch models to ONNX and TensorRT engines while preserving operator compatibility
- Increase FPS through INT8 post-training quantization calibration with minimal accuracy drop
- Eliminate CPU bottlenecks by fusing preprocessing (resize, normalise) and NMS into the GPU/NPU graph
- Package resilient 24/7 edge services with Docker, GStreamer and hardware video decoders
- Design telemetry, watchdogs and safe OTA model updates for intermittently connected field devices
Curriculum
Edge Hardware Ecosystem and Architecture Selection
- NVIDIA Jetson (Orin Nano/NX), Intel IPCs, Raspberry Pi + AI accelerators and mobile NPUs
- How TOPS, memory bandwidth, power envelope (Watts) and thermal limits shape model choice
Model Compilation with ONNX, TensorRT and OpenVINO
- Exporting from PyTorch to ONNX, dynamic axes, opset compatibility and graph fusion
- Layer fusion and hardware-specific kernel tuning with TensorRT and OpenVINO
FP16 and INT8 Quantization Calibration
- Preparing representative calibration datasets for Post-Training Quantization (PTQ)
- Layer-wise sensitivity analysis and Quantization-Aware Training (QAT) for detection and segmentation
End-to-End Camera Pipelines and Zero-Copy Memory
- Hardware-accelerated decoding from CSI and RTSP cameras with GStreamer
- Reducing CPU-GPU memory copies, on-GPU preprocessing and accelerated NMS
Field Operations: Packaging, Monitoring and OTA Updates
- L4T / JetPack containers, systemd watchdogs and offline buffering
- Device health telemetry (GPU/temperature/FPS), shadow testing and rollback-safe model updates
Other Courses in This Category
-
Knowledge distillation, structured pruning, Quantization-Aware Training (QAT), NAS and Torch-TensorRT profiling, for engineers shrinking and accelerating large computer vision models with minimal accuracy loss.
- Live Course
- 5 modules
-
Computer Vision
Comprehensive Computer Vision training ranging from classical image processing techniques to PyTorch-based CNN, Vision Transformer (ViT), Swin Transformer, and ConvNeXt architectures. Hands-on coverage includes object detection with YOLO and DETR, segmentation with SAM and U-Net, video analysis, 3D vision (NeRF, Gaussian Splatting), Diffusion models, and model deployment with ONNX/TensorRT.
- Live Course
- 13 modules
-
Digital image basics, filtering and feature extraction with OpenCV, the classification/detection/segmentation task types and a first prototype with pretrained models, for technical teams scoping computer vision use cases.
- Live Course
- 5 modules
-
CNN architectures (ResNet, EfficientNet, ConvNeXt) in PyTorch, transfer learning, data augmentation and training loops, for developers and data scientists who need to train and evaluate image models on company data.
- Live Course
- 5 modules
-
Self-supervised learning (DINOv2), vision foundation models (SAM 2, Grounding DINO), synthetic data, active learning and large-scale training, for experienced computer vision engineers tackling hard production problems.
- Live Course
- 5 modules
-
Object detection and instance segmentation with YOLO, RT-DETR and Mask R-CNN: labelling, training, mAP-based evaluation and tracking integration, for teams building an end-to-end detection/segmentation prototype.
- Live Course
- 5 modules
Get in touch
