New Release: Ultralytics v8.3.100

Discussion
,
1

:rocket: Announcing Ultralytics v8.3.100 Release!

We’re thrilled to share the release of Ultralytics v8.3.100, packed with exciting updates designed to improve compatibility, ease of use, and overall performance. This version focuses on enhancing PaddlePaddle support, updating documentation, streamlining model usability, and more. Here’s a quick dive into what’s new!

:glowing_star: Summary

The v8.3.100 release brings:

  • Advanced PaddlePaddle integration for accelerated workflows.
  • Updated pre-trained model links for YOLOv8, YOLOv9, YOLOv10, YOLO-NAS, YOLOE, and others.
  • Comprehensive improvements to YOLOE models, optimizing task and prompt management.
  • Fresh and enhanced documentation, including new tutorials and deployment guides.

:bar_chart: Key Changes

:rocket: PaddlePaddle Integration

  • Enhanced compatibility with PaddlePaddle >=3.0.0 for both CPU and GPU environments.
  • Improved model export: Now supports .json and .pdiparams formats.
  • Disabled OBB inference for Paddle due to reported silent errors causing mAP performance drops.

:books: Documentation Updates

  • Added a video tutorial for YOLO11 and NVIDIA DeepStream workflows.
  • Improved deployment guides for YOLO11 on NVIDIA Jetson devices.
  • Introduced detailed references for YOLOE usage, including visual prompts and integration with models like CLIP and MobileCLIP.

:building_construction: Pre-trained Model Updates

  • Updated pre-trained model links for consistent compatibility with v8.3.0.

:wrench: YOLOE Model Improvements

  • Visual Prompts: Changes prevent unintentional modifications to the original prompts during predictions.
  • Task Management: Automatic task detection for segmentation and detection tasks based on the Predictor, simplifying workflows.

:bullseye: Purpose & Impact

Purpose

  1. Strengthen PaddlePaddle compatibility and export reliability.
  2. Simplify model deployment with detailed guides and tutorials.
  3. Improve stability and usability for YOLOE models in advanced workflows.

Impact

  • Improved Compatibility: Leverage the power of PaddlePaddle >=3.0.0 seamlessly.
  • Enhanced Documentation: Clearer instructions for deploying YOLO models.
  • Streamlined Usability: Reduce errors and enhance efficiency with smarter YOLOE enhancements.
  • User-Friendly Resources: New video tutorials empower users to confidently use YOLO11 and deploy on NVIDIA platforms.

This release embodies our continued commitment to improving user experience, compatibility, and functionality across the suite of Ultralytics tools!

:link: What’s Changed

For all changes, explore the full changelog.

:speech_balloon: Get Involved

We’d love to hear from you! Download the new release, try out the features, and share your feedback with us. Your insights help shape the future of Ultralytics tools and the YOLO ecosystem.

:link: Download the Release
Let’s continue pushing the boundaries of AI together. Thank you for being part of the YOLO journey! :rocket: