ultralytics 8.3.206

v8.3.206 - `ultralytics 8.3.206` Support dynamic batch and image size with CoreML export (#22300)

on Python PyPI

🌟 Summary

CoreML export gets a big upgrade: dynamic image shapes and multi-image batching now work end-to-end, making Apple deployments more flexible and production-friendly. 🚀

📊 Key Changes

• CoreML dynamic shapes and batching
  • CoreML export now supports dynamic=True for variable image sizes and batch sizes. 🧩
  • Export validates safe combos: disallows dynamic with nms, and classification models can’t use dynamic.
  • Runtime updated to handle dynamic CoreML inputs correctly in prediction.
  • Tests expanded to cover dynamic CoreML scenarios and exclude invalid settings.
  • Docs export table updated to list dynamic for CoreML.
• Developer experience
  • Type hints added for lazily loaded classes (e.g., YOLO, YOLOWorld, YOLOE, SAM, etc.) via TYPE_CHECKING for better IDE autocomplete and static analysis. ✨ (@Y-T-G)
• Visualization robustness
  • plot_images now gracefully handles 1-, 2-, 3-, and multi-channel inputs (auto-pads or crops to 3 channels), improving reliability for multispectral and non-RGB data. 🖼️ (@ambitious-octopus)
• Documentation quality and speed
  • Many images migrated to Ultralytics-hosted AVIF for faster loads and fewer broken links; minor cleanup in Roboflow and integration guides. ⚡ (@glenn-jocher)
  • Predict docs now include tabbed Python and CLI examples for easier onboarding. 🧭 (@banu4prasad)
  • Fixed import path in Python docs to avoid confusion and errors. ✅ (@cdeil)
  • Small mkdocs config tidy-up to remove a duplicated setting. 🧹

🎯 Purpose & Impact

• Flexible Apple deployments
  • Run CoreML models with varying input sizes and multiple images per call, improving throughput and simplifying pipelines on macOS/iOS.
  • Removes a top pain point where batch inference worked for PyTorch but not CoreML.
• Safer, clearer exports
  • Better validation and warnings reduce misconfigurations (e.g., dynamic + nms) and save debugging time.
• Smoother dev workflow
  • Stronger type hints enhance editor support without changing runtime behavior.
• More robust training/previewing
  • Reliable plotting for 2-channel and multispectral images avoids visualization crashes and streamlines dataset exploration.
• Faster, clearer docs
  • AVIF assets speed up documentation; CLI snippets for predict reduce guesswork and help new users quickly run YOLO11.

Quick start examples:

# Export CoreML with dynamic shapes and set a max batch
yolo export format=coreml dynamic batch=16

# Batch inference with a CoreML package
yolo predict model=yolo11n.mlpackage batch=2

# Change image size on the fly (still batched)
yolo predict model=yolo11n.mlpackage imgsz=256 batch=2

Notes:

• For CoreML export: use dynamic=True for batch > 1.
• Don’t combine dynamic=True with nms=True.
• Classification models don’t support dynamic=True on CoreML.

What's Changed

• Fix import typo in python.md Docs by @cdeil in #22285
• GIF to AVIF conversion, smaller files size and same animation by @RizwanMunawar in #22293
• Enable type hints for lazy loaded classes by @Y-T-G in #22297
• Update YOLOE Docs image by @glenn-jocher in #22299
• Optimize images and update docs by @glenn-jocher in #22301
• Optimize SAM2 and ONNX images and update docs by @glenn-jocher in #22302
• Fix 2 channel TIFF support for plot_images by @ambitious-octopus in #22282
• Fix duplicate ignore_init_summary in mkdocs.yml by @glenn-jocher in #22305
• Add Python and CLI Examples to Predict Docs by @banu4prasad in #22303
• ultralytics 8.3.206 Support dynamic batch and image size with CoreML export by @Y-T-G in #22300

New Contributors

• @cdeil made their first contribution in #22285

Full Changelog: v8.3.205...v8.3.206