New Release: Ultralytics v8.3.93

Discussion
,
1

:tada: Announcing Ultralytics Release v8.3.93!

We’re excited to announce the release of Ultralytics v8.3.93! This update brings critical fixes, cutting-edge features, and documentation enhancements aimed at providing a smoother and more robust experience for the YOLO community. Here’s a detailed breakdown of what’s new in this release:

:glowing_star: Summary

Version 8.3.93 focuses on improving model robustness with a critical fix for TorchScript models, the introduction of YOLOE (a new real-time open-vocabulary detection and segmentation model), and several usability and performance enhancements for our documentation.

:rocket: New Features

YOLOE - Open-Vocabulary Detection and Segmentation

  • What it does: Detects unseen object classes using text, visual, and internal prompts. This brings YOLO into open-world detection scenarios with real-time performance!
  • Applications: Ideal for tasks like robotics, augmented reality (AR), and dynamic real-time applications requiring flexible object detection. :robot:
  • Visual Customization: Added txt_color support for annotation text color adjustments for better visual flexibility.

Stream Processing

  • Efficiency Boost: Introduced the stream functionality for memory-efficient video/image processing. This is especially useful for resource-constrained devices or high-resolution video inputs.

:hammer_and_wrench: Improvements

Documentation Enhancements

  • Performance Optimization:
    • Updated banner images to AVIF format, ensuring faster loading times.
    • Minified HTML, CSS, and JS files for improved performance and reliability.
  • Usability: Comprehensive documentation for YOLOE introduced to aid developers and users in open-vocabulary detection tasks. Parameter descriptions for tasks like export, prediction, validation, and visualization have been improved for clearer guidance.

:bug: Bug Fixes

TorchScript Models

  • Fix Description: Resolved an error when loading TorchScript models with Non-Maximum Suppression (NMS) by ensuring torchvision is pre-imported.
  • Impact: This ensures smooth deployments and prevents runtime errors, enhancing reliability for production environments.

:bullseye: Purpose and Impact

  • Model Reliability: The TorchScript fix guarantees seamless deployments across environments.
  • Cutting-Edge Features: YOLOE pioneers real-time open-vocabulary detection, pushing the boundaries of AI applications.
  • Enhanced Usability: Documentation optimizations improve accessibility, speed, and clarity for end users.

What’s Changed?

Full Changelog: See the changelog

:light_bulb: We Need Your Feedback!

We’d love to hear your thoughts on the new release! Try out the latest features, explore the new YOLOE functionality, and let us know your experience. Your feedback helps us refine and grow these tools for the entire YOLO community.

:inbox_tray: Download the release: Ultralytics Release v8.3.93

Thank you for continuing to support Ultralytics! This release is a testament to the collaborative spirit of the YOLO community—your contributions help make every update better.

Happy experimenting! :rocket: