Announcing YOLOE Release: Ultralytics v8.3.101
We’re excited to announce the release of Ultralytics v8.3.101, introducing major enhancements to the YOLOE model! This update focuses on simplifying video/stream handling, adding flexible inference options, improving export compatibility, and refining our documentation to boost ease of use.
Summary
This version shines with automated video handling, a rectangular inference option, and documentation fixes for seamless adoption. Community contributions have greatly enriched this release—thank you for driving innovation together!
Key Changes
Improved Video Handling:
-
Automatically selects the first frame of a video/stream as the visual prompt reference image, saving time for users.
-
Users can also specify a custom frame for better control.
Authored by @Laughing-q. For details, check the PR here.
New Rectangular Inference Option:
-
Added
rectparameter in predict mode to toggle rectangular padding during predictions. -
Enhances flexibility for unique datasets and scenarios requiring non-standard padding setups.
Contributed by @Laughing-q. See the PR here.
Detailed Documentation Updates:
-
Errors in YOLOE examples were fixed for clarity and correctness.
-
Bounding box formatting was clarified for ease of use.
-
Outdated integration notes were removed to reflect the latest features.
Special thanks to @JShengP, @RizwanMunawar, and @Y-T-G for these improvements. Check their combined efforts:
Model Export Compatibility:
-
Enhanced
class positional embeddingsduring export, offering smoother transitions and improved usability across environments. -
Fixes to Linear Probing examples simplify experimentation in advanced applications.
Contributions from @Y-T-G. Relevant PRs:
Purpose & Impact
This release ensures:
- Simpler workflows with automated first-frame selection for video/stream inputs.
- Flexible scenarios for inference through rectangular padding options.
- Straightforward onboarding, thanks to clearer documentation and fixed examples.
- Export reliability for advanced workflows.
- Stronger collaboration with gratitude for our contributors—thank you for making YOLOE even better!
These improvements target both experienced developers and newcomers, ensuring YOLOE is intuitive and adaptable to a wide range of use cases.
What’s Changed
Here’s a complete list of the updates included in v8.3.101:
- YOLOE Documentation and Visual Prompt Fixes: @JShengP, @RizwanMunawar, PR #19981, PR #19982.
- Dynamic Export Simplification: @Y-T-G, PR #19988.
rectOption for Predictions: @Laughing-q, PR #19994.- Linear Probing Example Fix: @Y-T-G, PR #19985.
- Visual Prompt Fix for Video Sources: @Laughing-q, PR #19959.
View the full changelog: v8.3.100…v8.3.101.
Release details, including direct downloads: Ultralytics v8.3.101 Release.
Try It Out!
We encourage you to update to v8.3.101 and explore these exciting enhancements. Your feedback is instrumental—please share your thoughts or report any issues via our GitHub Discussions.
Let’s keep pushing the boundaries of computer vision, together!
Thank you to all the contributors, users, and community members for your continued support—this release belongs to all of you. 
Stay innovative,
The Ultralytics Team