Ultralytics YOLO

🌟 Summary

Adds confidence scores to classification validation plots and strengthens NaN recovery during training, plus documentation updates highlighting SAM 3 and YOLO26. Cleaner experiment logs and more robust training—no breaking changes. 🚀

📊 Key Changes

• Classification visuals
  • Show top‑1 confidence on classification prediction plots during validation for clearer reads and quicker triage. See Show confidence in classification plots (PR #22365).
• Training stability and recovery
  • Ensure model is in train mode before NaN recovery checkpoint load to prevent inference‑tensor errors. See Train mode fix for NaN handling (PR #22381).
  • Recreate EMA before loading EMA state to avoid InferenceTensor errors on resume. See Recreate EMA for NaN recovery (PR #22372).
  • Check self.loss (not self.tloss) and refine NaN/fitness collapse detection logic for more reliable recovery. See Improve NaN handler loss check (PR #22382).
• Experiment logging
  • Reset results.csv when not resuming so new runs don't append to old logs with exist_ok=True. See Reset results.csv behavior (PR #22364).
• Documentation and navigation
  • New SAM 3 docs page (Coming Soon), YOLO26 marked "Coming Soon," and YOLO11 label cleanup for consistency; updated site navigation. See SAM 3 docs and model visibility updates (PR #22373).

🎯 Purpose & Impact

• Clearer model evaluation
  • Confidence overlays in classification validation plots help spot uncertain predictions and speed up debugging. 📈🔍
• More reliable training
  • Robust NaN/Inf handling reduces crashes and auto-recovers cleanly across YOLO11/YOLO26 and other models—no user action required. 🛡️
• Cleaner experiment management
  • Fresh results.csv per run (unless resuming) keeps metrics tidy and comparable. 🧹
• Better roadmap visibility
  • SAM 3 and YOLO26 appear clearly as "Coming Soon," helping users plan ahead while keeping YOLO11 docs consistent. 🧭

Tip: Upgrade with pip install -U ultralytics to get these improvements. ✨

What's Changed

• Reset results.csv if not resuming by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22364
• SAM 3 Docs page by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/22373
• Recreate EMA to avoid InferenceTensor error during NaN recovery by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22372
• Train mode fix for NaN handling by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/22381
• NaN handler check self.loss instead of self.tloss by @glenn-jocher in https://github.com/ultralytics/ultralytics/pull/22382
• ultralytics 8.3.214 Show confidence in classification plots by @Y-T-G in https://github.com/ultralytics/ultralytics/pull/22365

Full Changelog: https://github.com/ultralytics/ultralytics/compare/v8.3.213...v8.3.214