New Release: Ultralytics v8.3.112

Discussion
,
1

:rocket: Ultralytics v8.3.112 Release – Full Multispectral Support, COCO8-Multispectral, and More!

:glowing_star: Summary

We’re excited to announce Ultralytics v8.3.112! This release brings built-in multispectral image support across the entire Ultralytics ecosystem, including all YOLO tasks—training, validation, prediction, and export—not just for standard 3-channel (RGB) images, but for any number of channels (e.g., 10-channel multispectral imagery)! Plus, we’ve added a new COCO8-Multispectral dataset and a host of usability improvements.

Whether you’re working in remote sensing, agriculture, scientific research, or beyond, your multispectral workflows just became much easier.

:new_button: New Features

Multispectral (Multi-Channel) Image Support

  • Now supports images with any number of channels.
    YOLO models handle datasets with, for example, 10-channel TIFF images—right out of the box! This covers detection, segmentation, pose, classification, and more.
  • Dataset configurations now include a channels field.
  • Data loading, caching, and image reading robustly support multi-channel formats, including TIFF.
  • All workflows (train/val/predict/export) automatically use the correct channel count.

→ See details in New YOLO Multispectral Image Support by @Laughing-q

New COCO8-Multispectral Dataset

  • Added a lightweight, 10-channel version of the COCO8 dataset to help you quickly test multispectral pipelines.
  • Utility for converting RGB images to multispectral for experimentation.
  • Comprehensive documentation and usage guides included.

→ Explore the dataset in COCO8-Multispectral dataset PR by @glenn-jocher

:high_voltage: Improvements

  • Augmentations & Preprocessing:
    Data augmentations now intelligently adapt to the number of channels, applying only compatible transforms.
  • Visualization Upgrades:
    Plots and visual outputs now gracefully handle multi-channel data.
  • MobileSAM Documentation:
    Much clearer guidance plus improved YOLO comparison, making it easier to choose the right workflow for your needs.
    (PR by @glenn-jocher)
  • Callbacks Documentation:
    New, more relevant tutorial linked for easier onboarding.
    (PR by @RizwanMunawar)
  • ARM64 & Cross-Platform:
    Expanded test coverage ensures reliability on more systems.
    (ARM64 Docker images test by @lakshanthad)
  • Classification Dataset Splitting:
    Enhanced tools and docs for splitting classification datasets.
    (Refactor by @glenn-jocher)
  • Logging Improvements:
    Warning and error messages now use clear prefixes for better diagnostics.
    (Logger update by @glenn-jocher)
  • General Robustness:
    Numerous tweaks to streamline testing, error handling, and code clarity.

:lady_beetle: Bug Fixes and Maintenance

  • Fixed an issue with stale workflow runs (PR by @Y-T-G)
  • Fixed test_solutions.py Streamlit test failures (PR by @glenn-jocher)
  • Allowed validation with rect argument for dynamic models (PR by @Y-T-G)
  • Fixed Annotator to ensure list handling and resolved classification results background transparency (PRs by @glenn-jocher)

:bullseye: Why This Matters

  • Advanced Image Analysis: Researchers and practitioners can now work with real multispectral and hyperspectral imagery across all core Ultralytics tools.
  • No Hacks Needed: All common and advanced use cases for multi-channel data are directly supported—no workarounds.
  • Ready Out of the Box: Start using multispectral imagery with existing pipelines using the new COCO8-Multispectral dataset or your own data.

:link: Resources

:raising_hands: Try It Out & Share Your Feedback!

Update to v8.3.112 and explore multispectral workflows today!
We love hearing from our community—please share your experience, suggestions, or any issues you encounter in the Ultralytics Discussions or directly by opening an issue.

The Ultralytics team and the amazing YOLO community are grateful for your continued support and contributions. Let’s keep building the future of open computer vision—together!