🌟 Summary
This release enhances support for Apple's MPS devices and includes various improvements for better performance and usability across platforms.
📊 Key Changes
- Enhanced memory logging for Apple's MPS on macOS, ensuring accurate GPU memory usage reporting.
- Improved prediction handling in the model, including saving predictions and consistent bounding box handling.
- Updated documentation for seamless navigation and improved clarity.
- Fixed issues with
mps.empty_cache()for macOS without MPS, preventing crashes. - Introduced new benchmarks for Intel's latest hardware, providing key performance metrics.
🎯 Purpose & Impact
- 🖥️ macOS Improvements: Accurate memory reporting and management enhance training and inference performance on macOS, especially for Apple hardware users.
- 📈 Performance Benchmarks: Intel users gain insights into the performance of new hardware, helping them optimize their setups.
- 📄 Documentation and Usability: Updated docs enhance user experience by making it easier to find information and contribute, fostering community growth.
- 📐 Enhanced Prediction Handling: Saving predictions and improving mask-less bounding box handling enhance the output process, making models more robust and user-friendly.
- 🚀 Optimized Platform Support: The changes ensure smoother operation across different environments, helping all users maximize the capabilities of Ultralytics models.
What's Changed
- Return boxes for SAM prompts inference by @Laughing-q in #16276
- Docs improvements and redirect fixes by @glenn-jocher in #16287
- Fix
mps.empty_cache()for macOS without MPS by @Skillnoob in #16280 - Add color palette tables to docs by @jk4e in #16242
- Intel Core Ultra benchmarks by @ambitious-octopus in #15895
ultralytics 8.2.94Apple MPS train memory display by @Oil3 in #16272
Full Changelog: v8.2.93...v8.2.94