Ultralytics YOLO

🌟 Summary (single-line synopsis) Ultralytics 8.3.249 adds an official NVIDIA ARM64 Docker image for running YOLO11 on JetPack 7 / ARM64 GPUs (plus a few usability + contributor/docs improvements) 🚀🐳 📊 Key Changes 🐳 New Docker image for NVIDIA ARM64 (JetPack 7 / ARM64 GPUs) Added docker/Dockerfile-nvidia-arm64 based on NVIDIA's PyTorch NGC image (nvcr.io/nvidia/pytorch:25.10-py3) CI now builds & publishes ultralytics/ultralytics:latest-nvidia-arm64 Container tweaks for smoother out-of-the-box use: Uses opencv-python-headless (avoids GUI dependency issues in containers) 🧩 Installs ARM64 onnxruntime_gpu + reinstalls torch/torchvision for CUDA compatibility 🔥 Preloads Ultralytics fonts + includes a default YOLO11 model (yolo11n.pt) 📦 JetPack 7 docs updated with pull/run commands ✅ 🔗 Support for ul:// URIs (Ultralytics Platform / HUB-style referencing) New resolve_platform_uri() resolves ul://... into signed download URLs (using ULTRALYTICS_API_KEY) check_file() now accepts ul://... and automatically downloads/caches the file locally 📥🗂️ Better error messages for invalid keys, forbidden access, not found, and "still processing" cases 🚦 🧮 Fix: ObjectCounter label display logic Corrected operator precedence in ObjectCounter.display_counts so show_in / show_out flags behave correctly (prevents empty/incorrect labels) 🛠️ 📝 Contributor workflow improvements Updated CONTRIBUTING.md with PR size/scope guidance, feature PR expectations, and automated review notes to streamline contributions 🤝 🎯 Purpose & Impact 🚀 Easier deployment on new NVIDIA ARM64 platforms (Jetson AGX Thor, DGX Spark, JetPack 7) Reduces setup friction (drivers/libs/Python deps) and gets users to a working YOLO11 environment faster—especially useful for edge + embedded deployments. 🐳 More consistent "it just works" container experience Headless OpenCV and pinned runtime dependencies lower the chance of common container runtime/import issues, improving reliability for production and CI. 🔗 Simpler sharing & reproducibility with ul:// Users can reference datasets/models by a short URI and let Ultralytics handle secure resolution + download, making scripts and configs cleaner and more portable. ✅ Cleaner analytics overlays in ObjectCounter Prevents misleading or empty class labels when only IN/OUT is enabled—better visual outputs with fewer surprises. 📘 Faster, clearer open-source contributions Better guidelines help contributors submit reviewable PRs, which can speed up merges and reduce back-and-forth. If you want to try the new ARM64 image on JetPack 7: t=ultralytics/ultralytics:latest-nvidia-arm64 sudo docker pull $t && sudo docker run -it --ipc=host --runtime=nvidia $t What's Changed Add scope and PR size guidelines to CONTRIBUTING.md by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/23133 Fix typo in CONTRIBUTING.md guidelines by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/23134 Fix operator precedence in ObjectCounter.display_counts by @raimbekovm in https://github.com/ultralytics/ultralytics/pull/23132 URI resolution by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/23136 ultralytics 8.3.249 Jetson AGX Thor and DGX Spark Docker by @lakshanthad in https://github.com/ultralytics/ultralytics/pull/23111 Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.248...v8.3.249