New Release: Ultralytics v8.3.225

[glenn-jocher]

Ultralytics v8.3.225 is out

Summary
Ultralytics 8.3.225 introduces first-class KITTI dataset support for YOLO11, safer checkpoint loading for SAM/SAM2, a breaking change for SAM-2 object IDs (now 0-based), and more reliable Jetson JetPack 4 Docker builds, alongside polished docs and type hints. If you’re starting with YOLO today, YOLO11 is our recommended default for all use cases.

New Features

• KITTI dataset support for YOLO11
  Train and evaluate out-of-the-box with a ready-to-use kitti.yaml, docs, and examples. Explore the new KITTI dataset documentation and review the kitti.yaml configuration.

Improvements

• Safer checkpoint loading for SAM/SAM2
  A unified torch_load helper improves cross-version PyTorch compatibility and avoids pickle pitfalls.
• Jetson JetPack 4 Docker build reliability
  Builds are now more reproducible by sourcing ONNX Runtime/TensorRT wheels from Ultralytics-hosted assets.
• Docs and developer experience polish
  Cleaned code blocks for ExecuTorch docs, stronger type hints across dataloaders and utilities, and CI stability improvements.

Breaking Changes

• SAM-2 interactive predictor uses 0-based object IDs
  This standardizes behavior with Python conventions and helps prevent off-by-one errors. If your scripts assumed IDs starting at 1, please update them to start at 0.

Why this matters

• Faster starts on autonomous driving research
  KITTI support lets you get training and benchmarking with YOLO11 in minutes.
• More reliable model loading across environments
  Centralized checkpoint handling reduces surprises between PyTorch versions.
• Reproducible hardware builds
  Jetson images are less likely to break due to changing external assets.
• Smoother onboarding
  Clearer docs and typing help IDEs and new contributors.

Get started on KITTI with YOLO11

pip install -U ultralytics

yolo detect train data=kitti.yaml model=yolo11n.pt epochs=100 imgsz=640

from ultralytics import YOLO

model = YOLO("yolo11n.pt")
model.train(data="kitti.yaml", epochs=100, imgsz=640)

Merged Pull Requests

• PR #22576: Fix 429 forum links in Dockerfile-jetson-jetpack4 by @glenn-jocher
• PR #22577: Use torch_load patch for weights_only cases by @Laughing-q
• PR #22581: Fix indentation in integrations/executorch.md examples by @Laughing-q
• PR #22582: Remove redundant Executorch test by @Laughing-q
• PR #22580: Add Python type hints for dataloader functions by @Laughing-q
• PR #22529: Fix object ID indexing with SAM2DynamicInteractivePredictor by @Y-T-G
• PR #22539: Add KITTI dataset (ultralytics 8.3.225) by @RizwanMunawar

Release and changelog

• Read the detailed notes on the Ultralytics v8.3.225 release page.
• Review every change in the full v8.3.224 → v8.3.225 comparison.

We’d love your feedback
Give the release a try, especially the new KITTI support and SAM/SAM2 improvements, and let us know how it goes in Discussions or by opening an issue. Your input helps the YOLO community and Ultralytics team keep improving.