senior-computer-vision

To deploy a PyTorch object detection model to edge hardware, this skill guides you through dataset preparation, training, quantization, and runtime conversion to achieve sub-30ms inference latency using optimized formats like ONNX and TensorRT.

The best way to optimize YOLOv8 models for production inference involves automated benchmarking and conversion to high-performance runtimes like TensorRT, OpenVINO, and ONNX, which this skill handles through the Ultralytics and PyTorch ecosystem.

Yes, you can use this computer vision pipeline for instance segmentation and image classification. It provides end-to-end support for YOLO, Faster R-CNN, and SAM architectures across the full model lifecycle from dataset preparation to deployment.

Yes, you need ONNX runtime, OpenCV, PyTorch, torchvision, Ultralytics, and timm installed. These dependencies are required to execute the training, model optimization, and inference workflows for building production-grade computer vision systems.

High inference latency in real-time object detection often occurs when models are not converted to optimized runtimes. This skill automates quantization and conversion to TensorRT, OpenVINO, and ONNX to ensure sub-30ms latency on edge or cloud hardware.

Yes, this skill supports model conversion to TensorRT and OpenVINO runtimes. It provides automated benchmarking and optimization workflows to convert PyTorch and Ultralytics models into high-performance inference formats for edge and cloud deployment.