ultralytics 8.3.152

v8.3.152 - `ultralytics 8.3.152` Optimize padding for Segment mask processing (#20957)

on Python PyPI

🌟 Summary

Enhanced segmentation mask accuracy, improved memory management, faster confusion matrix processing, and better documentation for Ultralytics YOLO11. 🚀🖼️

📊 Key Changes

• Segmentation Mask Precision: Optimized how padding and cropping are handled when resizing segmentation masks, ensuring masks align perfectly with original images. (PR #20957)
• Model Task Loading: Improved how the model’s task type is determined when loading checkpoints, making model loading more reliable. (PR #20966)
• Memory Management: Automatically clears GPU memory before validation if VRAM usage is high, helping prevent out-of-memory errors during training. (PR #20960)
• Confusion Matrix Speed: Made confusion matrix calculations up to 30% faster by optimizing how data is processed. (PR #20972)
• Confusion Matrix Readability: Enhanced the contrast of text in confusion matrix plots for easier interpretation. (PR #20955)
• Custom Classification Augmentations: Added clear documentation and examples for customizing data augmentations in classification tasks. (PR #20949)
• Video Tutorials in Docs: Embedded new and updated YouTube tutorials for training on Objects365 and for workout monitoring, making it easier to learn by example. (PR #20965)

🎯 Purpose & Impact

• More Accurate Results: Segmentation users will see improved mask alignment, leading to more precise object boundaries and better visual results.
• Fewer Training Interruptions: Automatic VRAM management reduces the risk of training crashes due to memory spikes, especially on limited hardware.
• Faster Evaluation: Speedier confusion matrix processing means quicker feedback during model evaluation, saving time for all users.
• Easier Model Loading: Models now retain their correct task type when loaded, reducing errors and confusion.
• Better Learning Resources: New video guides and clearer documentation help both beginners and advanced users get the most out of Ultralytics YOLO11.
• Enhanced Usability: Improved plot readability and flexible augmentation options make the toolkit more user-friendly and adaptable to custom needs.

Overall, this release delivers a smoother, more reliable, and more accessible experience for all Ultralytics YOLO11 users! 🎉

What's Changed

• Fix incorrect task when loading saved checkpoint by @Y-T-G in #20966
• Add https://youtu.be/J-RH22rwx1A and https://youtu.be/Ck7DW96dNok to docs by @RizwanMunawar in #20965
• Clear memory if VRAM usage is high before validation. by @RaahimSiddiqi in #20960
• Optimize ConfusionMatrix process_batch performance by @dianyo in #20972
• Improve text coloring across normalize modes for confusion matrix (≤30 classes) by @mihlefeld in #20955
• Add custom example for classification augmentation by @Laughing-q in #20949
• ultralytics 8.3.152 Optimize padding for Segment mask processing by @horsto in #20957

New Contributors

• @horsto made their first contribution in #20957
• @mihlefeld made their first contribution in #20955
• @RaahimSiddiqi made their first contribution in #20960

Full Changelog: v8.3.151...v8.3.152