New Release: Ultralytics v8.3.118

:rocket: Ultralytics v8.3.118 Release: MobileCLIP, Grayscale Support & Improved Compatibility!

:glowing_star: Summary

We’re excited to announce Ultralytics v8.3.118! This release introduces support for the new TorchScript MobileCLIP model, expands grayscale image handling, and strengthens device and package compatibility with broader testing—all while keeping your experience modern and streamlined.

Whether you’re working with vision-language applications, medical imaging, or edge devices, this release has meaningful updates for you!


:sparkles: New Features

TorchScript MobileCLIP Integration :mobile_phone:

  • Added support for Apple’s TorchScript MobileCLIP, allowing fast and portable text encoding for vision-language applications.
  • Enhances custom class naming and improves performance on mobile and edge hardware.
  • PR 20354 by @Y-T-G

Grayscale Image Training Support :black_heart:

  • Out-of-the-box support for grayscale datasets and augmentations.
  • Simplifies training/inference on medical images, document scans, and more—no extra preprocessing required.
  • PR 20310 by @Laughing-q

:hammer_and_wrench: Improvements

Expanded Continuous Integration (CI) Coverage :test_tube:

  • Raspberry Pi and Conda environment CI tests now run by default, boosting cross-platform reliability.
  • Test files included in Conda builds for higher package quality.
  • Ensures YOLO models are ready-to-roll on a wide range of devices.
  • PR 20365 by @glenn-jocher
  • PR 20361 by @glenn-jocher

YOLO11 as Default Reference :package:

  • All documentation, code examples, and Dockerfiles now reference YOLO11.
  • Helps guide users to the latest state-of-the-art.
  • PR 20367 by @glenn-jocher

Cleaner CoreML Handling :broom:

  • Legacy CoreML and protobuf workarounds removed.
  • Streamlined codebase for easier maintenance and improved compatibility moving forward.
  • PR 20362 by @glenn-jocher

:lady_beetle: Bug Fixes & Housekeeping

  • Various behind-the-scenes improvements for device robustness, test coverage, and easier deployment—especially for Conda and edge hardware.

:chart_increasing: Full Changelog & How to Upgrade

See the full details in the
Ultralytics v8.3.118 Release Notes
or compare changes at
Changelog between v8.3.117 and v8.3.118.

Upgrade today and let us know how it works for your projects!

pip install -U ultralytics

:speech_balloon: Try It Out & Join the Conversation

We value your feedback—Ultralytics releases are shaped by real-world use and community input. If you have ideas, questions, or run into issues, please share your thoughts right here in the forum.

Thank you for being part of the YOLO community and helping us push the boundaries of open-source vision AI!

— The Ultralytics Team