New Release: Ultralytics v8.3.175

Discussion
, ,
1

:rocket: Ultralytics v8.3.175 Release: Safer Deployments & Smoother Workflows

:glowing_star: Quick Summary

We’re excited to announce the release of Ultralytics v8.3.175! This update focuses on safer model deployment, enriched documentation, clearer results, expanded export testing, and improved notebooks—all designed to make your experience more robust and enjoyable.

If you’ve been waiting for a release that prioritizes model safety, ease of use, and reliability in both training and deployment, this is it!

:sparkles: New Features & Major Improvements

Model Safety for YOLOE

Preventing post-fusion class changes in YOLOE raises production safety. Once your model is fused for inference, its classes can’t be accidentally altered—making your deployments more reliable.

Expanded YOLOE Documentation

We’ve added step-by-step guides covering fine-tuning and linear probing for both segmentation and detection, empowering you to train custom YOLOE models with confidence.

Enhanced Logging in Cropping Solutions

Logs now include detected object classes and their counts—making results more transparent and easier to debug or interpret.

Comprehensive Export Testing

Significant expansion of NCNN export tests means your models are more likely to run smoothly across different environments and setups.

Example Notebooks—Now Even Better

All Colab and example notebooks have been updated to the latest Ultralytics release, featuring refreshed outputs and improved user tips to help you get started with less friction.

:hammer_and_wrench: Improvements & Bug Fixes

  • Batch Size Consistency for TensorRT:
    Validation and inference with TensorRT-exported models now handle batch sizes more robustly.
    Fix by @syedhamzamohiuddin
  • Queue Management Docs Update:
    Quick-launch Colab badge for the queue management notebook makes resource handling a breeze.
    Docs update by @RizwanMunawar

:compass: Purpose & Impact

  • Safer Model Deployment: Protects against accidental model misconfigurations after optimization.
  • Easier Custom Training: New documentation guides simplify the process of adapting YOLOE for your data and tasks.
  • Clearer Results & Debugging: Enhanced logging lets you see exactly what your models detect and how many.
  • Increased Export Reliability: More thorough export tests minimize surprises in production.
  • Lowered Barriers for New Users: Badges and notebook upgrades make experimenting smooth and accessible.

:new_button: Notable PRs & Contributors

Here’s a roundup of what’s changed in v8.3.175:

A special welcome and thanks to @syedhamzamohiuddin for their first contribution!

:vertical_traffic_light: Ready to Try?

Upgrade with pip install -U ultralytics and explore the latest features! Your feedback is always appreciated—let us know what works, what could be better, and any successes or issues you encounter.

Release page: Ultralytics v8.3.175 Release

Thanks for being part of the YOLO community and helping us advance open-source vision together!
— The Ultralytics Team