Ultralytics v8.3.249 is out! 
(YOLO11 on NVIDIA ARM64 / JetPack 7)
TL;DR: Ultralytics v8.3.249 ships an official NVIDIA ARM64 Docker image for running Ultralytics YOLO (YOLO11) on JetPack 7 / ARM64 GPUs, plus ul:// URI support, a small analytics overlay fix, and contributor/docs improvements. ![]()
New Feature: Official NVIDIA ARM64 Docker image (JetPack 7 / ARM64 GPUs)
If youâre deploying on Jetson AGX Thor, DGX Spark, or other JetPack 7 ARM64 platforms, this release adds a first-class container path:
- Added
docker/Dockerfile-nvidia-arm64based on NVIDIA PyTorch NGC (nvcr.io/nvidia/pytorch:25.10-py3) - CI now builds & publishes
ultralytics/ultralytics:latest-nvidia-arm64 - Container âit just worksâ improvements:
- Uses
opencv-python-headlessto avoid GUI dependency issues in containers
- Installs ARM64
onnxruntime_gpuand reinstallstorch/torchvisionfor CUDA compatibility
- Preloads Ultralytics fonts and includes a default YOLO11 model (
yolo11n.pt)
- Uses
- JetPack 7 docs updated with pull/run commands

Try it:
t=ultralytics/ultralytics:latest-nvidia-arm64
sudo docker pull $t && sudo docker run -it --ipc=host --runtime=nvidia $t
New Feature: ul:// URI support (Platform-style referencing)
You can now reference assets via ul://... and let Ultralytics handle secure resolution + download:
- New
resolve_platform_uri()resolvesul://...to signed download URLs (viaULTRALYTICS_API_KEY) check_file()now acceptsul://...and auto downloads/caches locally

- Clearer errors for invalid keys, forbidden access, not found, and âstill processingâ cases

Bug Fix: ObjectCounter label display logic
A small but important fix to analytics overlays:
- Corrected operator precedence in
ObjectCounter.display_countssoshow_in/show_outbehave properly (prevents empty/incorrect labels)
Contributor / Docs Improvements
Contributor workflow got clearer and more streamlined:
- Updated
CONTRIBUTING.mdwith PR size/scope guidance, feature PR expectations, and automated review notes
Whatâs Changed (PRs)
- Add scope and PR size guidelines to
CONTRIBUTING.mdby @Y-T-G in PR #23133 - Fix typo in CONTRIBUTING.md guidelines by @Y-T-G in PR #23134
- Fix operator precedence in
ObjectCounter.display_countsby @raimbekovm in PR #23132 - URI resolution by @glenn-jocher in PR #23136
ultralytics 8.3.249Jetson AGX Thor and DGX Spark Docker by @lakshanthad in PR #23111
Upgrade & Share Feedback
Upgrade with:
pip install -U ultralytics
You can review the full details in the v8.3.249 GitHub release, and browse the full changelog comparing v8.3.248âŚv8.3.249.
If you try the new ARM64 container (or the new ul:// flows), let us know what hardware youâre running and how it goesâfeedback helps us keep edge deployments smooth and reproducible.