ultralytics 8.3.197

v8.3.197 - `ultralytics 8.3.197` New Construction-PPE dataset (#21976)

on Python PyPI

🌟 Summary

Ultralytics 8.3.197 adds the new Construction-PPE dataset and delivers training, export, plotting, HUB, and docs improvements that make YOLO11 workflows more robust and easier to use. 🦺🚀

📊 Key Changes

• New dataset: Construction-PPE for safety compliance
  • Added dataset page with usage, YAML, samples, and FAQs for detecting worn and missing PPE such as helmets, vests, gloves, boots, and goggles. 🦺
  • See the Construction-PPE docs and the construction-ppe.yaml for quick starts with YOLO11.
• Training pipeline simplification and torch.compile compatibility
  • Decoupled forward and loss (preds = model(batch["img"]); loss = model.loss(batch, preds)), removed compile-disabling wrappers, and cleaned up dynamic tensor handling for detect/pose/segment. ⚙️
• More reliable exports for OBB and INT8
  • Added an exit_early switch to Fast-NMS and set exit_early=False during export to prevent tracing breaks and empty outputs in OBB and some INT8 backends. 📦
• Smarter segmentation plots
  • Validation plots now respect max_det instead of a fixed 50-mask cap, giving you consistent, configurable visualizations. 🖼️
• Ultralytics HUB logging fix
  • HUB session is now passed during Trainer initialization for more reliable logging and syncing throughout training. 📡
• Jetson CI maintenance
  • Net effect: Jetson JetPack 5 Docker builds are re-enabled in CI, restoring coverage for JP5 users. 🐳
• Documentation improvements
  • New COCO128-Seg dataset page for quick segmentation tests and debugging, plus navigation and broken link fixes. 📚

Helpful links:

• Construction-PPE dataset docs: Construction-PPE Dataset
• Dataset YAML: construction-ppe.yaml in the repository
• COCO128-Seg docs: COCO128-Seg Dataset
• Ultralytics HUB: Ultralytics HUB

🎯 Purpose & Impact

• Safety AI out-of-the-box: Construction-PPE enables rapid prototyping of real-time compliance monitoring and alerts on construction sites. 🏗️
• Faster, more stable training: Cleaner torch.compile integration benefits PyTorch 2.x users and improves performance, including on Jetson devices. ⚡
• Fewer export headaches: The Fast-NMS export tweak reduces empty detections and tracing errors for OBB and certain INT8 pipelines. ✅
• Clearer segmentation visuals: Plotting aligns with your max_det settings for more intuitive validation; tune max_det if plotting slows. 🎛️
• Reliable HUB workflows: Early, consistent session handling boosts logging and synchronization with Ultralytics HUB projects. 🔒
• Better onboarding: Expanded docs (COCO128-Seg, dataset indexes, authors) smooth out getting started and testing pipelines. 🧭

Enjoy the new dataset and smoother YOLO11 experience! 🎉

What's Changed

• Disable Jetson Jetpack 5 image for debugging by @glenn-jocher in #21989
• Fix metrics logging to HUB by @Y-T-G in #21993
• Fix empty predictions with OBB NMS export by @Y-T-G in #21971
• Increase instance plotting limit for segmentation to max_det by @Y-T-G in #21992
• Remove disable_dynamo wrapper around Detect.inference() and loss calculation by @Y-T-G in #21996
• Add COCO128-Seg dataset documentation and fix broken links by @fcakyon in #22004
• ultralytics 8.3.197 New Construction-PPE dataset by @RizwanMunawar in #21976

Full Changelog: v8.3.196...v8.3.197