Ultralytics v8.3.119 Release – CutMix Augmentation, Lighter Docker, Improved Export, and More!
Summary
We’re excited to announce Ultralytics v8.3.119! This update brings the powerful CutMix image augmentation, streamlined Docker and export workflows, flexible logging integrations, and smoother dependency management. It’s all about helping you build even more robust, production-ready models with enhanced ease and flexibility. Check out what’s new below and let us know how it works for you!
View the full release on GitHub
New Features
CutMix Data Augmentation
- What’s new: Introducing CutMix, an augmentation technique that randomly combines patches from two images (and their labels) during training. This helps models better handle occlusions—a valuable boost for real-world applications such as autonomous driving or crowded scene detection.
- Supported tasks: Detection and Classification.
- How to use: Configure the new
cutmixhyperparameter (probability) and optionally thebetamixing ratio. - Docs: Detailed usage and configuration now included in the augmentation guides and hyperparameter tables.
- PR: New CutMix image augmentation by @artzuros
Docker and Dependency Improvements
- Dockerfiles cleaned: Removed unnecessary packages (
g++,libusb-1.0-0,keras) making containers lighter and faster to build/deploy. - Documentation updated: The Docker Quickstart guide is now clearer and less redundant.
- PRs:
Export and Logging Enhancements
- ONNXSlim Dependency: Updated to latest version for more reliable ONNX and TensorFlow exports.
Update onnxslim>=0.1.46 by @glenn-jocher - TFLite Export: Process is now more robust and Python-version friendly. Removed legacy dependencies (
tflite_support,flatbuffers).
Remove tflite_support and flatbuffers by @glenn-jocher - Logging: Default integrations now with Weights & Biases and MLflow for seamless experiment tracking.
Add W&B and MLflow logging by @glenn-jocher
Documentation & Usability Updates
- CutMix: Fully documented in augmentation and hyperparameter guides.
- Label Format: Improved and clarified for easier development and extending.
- Contributors: New contributors recognized in documentation.
- PRs:
New Contributors
- @nikenake1: Improved box format comments
- @SHOscarChen: Docker Quickstart docs
- @artzuros: CutMix augmentation
Thanks to all contributors and our incredible community for your ongoing support and innovations!
Try It Now & Share Feedback
Upgrade with:
pip install --upgrade ultralytics
Or try out the latest release.
Explore CutMix, benefit from a smoother Docker/export experience, and enjoy enhanced tracking with W&B and MLflow! Your feedback, suggestions, and bug reports are invaluable—please join the conversation and help us continue to improve Ultralytics for everyone.
Full Changelog:
See all changes between v8.3.118 and v8.3.119
Happy training!
— The Ultralytics Team