jetson-inference

Why we included this project

If you build real-time vision systems on NVIDIA Jetson hardware, this is one of the most practical starting points available. It bundles TensorRT-optimized implementations of the core deep vision primitives, image classification, object detection, semantic segmentation, pose estimation, and action recognition, with C++ and Python APIs plus console and camera-based examples. Rather than just dropping in pretrained models, it walks you through the whole onboard workflow, from collecting your own datasets and fine-tuning with PyTorch to converting to TensorRT and deploying with low latency on embedded GPUs. That makes it especially useful for robotics engineers and embedded developers who need vision to run reliably at the edge off a live camera feed. The ROS/ROS2 support and WebRTC webapp examples also smooth the path from a working demo to a real deployed system.