New Release: Ultralytics v8.3.113

Discussion
,
1

:rocket: Ultralytics v8.3.113 Release: Multispectral Power & Enhanced Usability!

:glowing_star: Summary

Ultralytics v8.3.113 is out now, bringing groundbreaking multispectral dataset support, smarter multi-channel image processing, new device flexibility with Intel/OpenVINO, and a host of usability and visualization upgrades! This release continues our mission to make cutting-edge computer vision both advanced and accessible for all users.

We encourage everyone—researchers, practitioners, and enthusiasts alike—to try out these new features, explore the updated capabilities, and share feedback to help keep moving YOLO and Ultralytics forward!

:new_button: New Features

1. DOTA8-Multispectral Dataset Support

  • We’ve added built-in configuration for the DOTA8-Multispectral dataset, enabling seamless training and evaluation on 10-channel multispectral images. This unlocks new horizons for aerial, satellite, and remote sensing tasks!
    PR by @glenn-jocher

2. Improved Multispectral Conversion

  • The convert_to_multispectral utility has been upgraded, along with code examples, for easier creation and handling of 10-channel datasets.

3. Flexible Intel Device Selection (OpenVINO)

  • Users can now specify their Intel hardware (CPU, NPU, GPU) for inference with better automatic device detection and warnings if something goes wrong.
    PR by @ambitious-octopus

4. Flexible Validation with classes Arg

  • Added the ability to validate your models on specific classes using the new classes argument for finer-grained evaluation.
    PR by @Y-T-G

:hammer_and_wrench: Improvements

  • Robust Multi-Channel Image Handling: Enhanced processing logic to correctly handle grayscale, RGB, and multispectral images, minimizing errors across pipelines.
    PR by @Laughing-q

  • SAHI OBB Video Inference:
    More flexible device selection, new slice size options, and updated docs make video inference with SAHI and YOLO11 even easier.
    PR by @RizwanMunawar

  • Region Counting Visualization:
    Region-based counting now features improved color-coded tracking and larger, more readable overlays for easier interpretation.
    PR by @RizwanMunawar

  • Ultralytics Documentation Widget Upgrade:
    Our documentation site now includes a refreshed widget with sleek branding, dark-mode support, and one-click access to community resources.
    Widget upgrade: PR by @glenn-jocher
    Dark mode fixes: PR by @glenn-jocher

:lady_beetle: Bug Fixes & Reliability

  • Safer, more reliable handling of diverse channel formats—no more pipeline crashes due to unexpected image input types.
  • Better automatic device checks and error handling when using OpenVINO on various Intel hardware platforms.

:loudspeaker: Get Involved & Share Your Thoughts!

We invite you to:

  • Test the new release: Ultralytics v8.3.113 on GitHub
  • Review the full changelog for v8.3.113
  • Join the conversation below and let us know how the new multispectral and hardware features work for your projects
  • Report your experience, suggest improvements, or showcase what you’re building!

A huge thank you from the Ultralytics team and the wider YOLO community to all contributors and users helping move vision AI forward. Your feedback and collaboration drive these innovations!

Happy experimenting!
— The Ultralytics Team