New Release: Ultralytics v8.3.152

Discussion
,
1

:rocket: Ultralytics v8.3.152 Release Announcement

Hello YOLO community!

We’re excited to announce the release of Ultralytics v8.3.152, packed with enhancements focusing on segmentation accuracy, smarter memory use, smoother model loading, and even more user-friendly documentation. This update brings improvements that benefit everyone — from researchers and developers to first-time YOLO users!

:glowing_star: Quick Summary

  • Sharper segmentation masks & improved results
  • Automatic VRAM management for smoother training
  • Faster, clearer confusion matrix plots
  • Better docs with new YouTube tutorial content
  • More reliable checkpoint/model loading
  • Enhanced custom classification augmentation documentation

:new_button: New Features & Major Improvements

Enhanced Segmentation Mask Precision

  • Optimized resizing logic ensures segmentation masks align more accurately with original images for precise object boundaries.
  • Thanks to @horsto (PR #20957).

Smarter Model Loading

  • Improved task detection when loading checkpoints means your models retain the correct configuration, reducing confusion and setup errors.
  • Thanks to @Y-T-G (PR #20966).

Automatic GPU Memory Management

  • The framework automatically frees GPU memory before validation if VRAM usage is high, minimizing accidental out-of-memory errors—especially helpful for those with limited hardware.
  • Thanks to @RaahimSiddiqi (PR #20960).

:chart_increasing: Improvements

Faster & Clearer Confusion Matrix

  • Confusion matrix calculations are up to 30% faster, letting you evaluate results even more quickly.
  • Improved readability with higher-contrast text, so your evaluation visuals are easier to interpret.
  • Credits: @dianyo (PR #20972), @mihlefeld (PR #20955).

Better Docs, More Learning

  • New and updated YouTube tutorials embedded in the docs, covering topics like Objects365 and workout monitoring, so you can learn by example.
  • Guide updates by @RizwanMunawar (PR #20965).

Custom Classification Augmentations

  • New documentation and examples for customizing augmentations in classification tasks, so you can adapt workflows to your datasets.
  • Thanks to @Laughing-q (PR #20949).

:hammer_and_wrench: Bug Fixes

  • Several core logic improvements make YOLO11 training and inference more robust, minimizing interruptions and maximizing reliability.

:waving_hand: Welcome New Contributors!

A warm welcome to new contributors who made their first commits in this release:

Your contributions help move the whole community forward!

:link: Helpful Links

:triangular_flag: How to Try the New Release

Upgrade your Ultralytics package via pip:

pip install -U ultralytics

Explore the new features and improvements today! For detailed migration notes, see the full changelog.

:speech_balloon: We Value Your Feedback!

As always, your input is invaluable. Please post your experiences, bug reports, and suggestions in our GitHub Discussions or directly on the issue tracker. Together, we’ll keep pushing YOLO forward!

A huge thanks to everyone in the community and the Ultralytics team for your work, feedback, and support. Happy building and training! :rocket: