Update default Albumentations

[glenn-jocher]

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Optimized image augmentation probabilities in the YOLOv5 model.

📊 Key Changes

• Reduced the probability of applying the Blur and MedianBlur effects from 0.1 to 0.01.
• Added new augmentations with a probability of 0.01 for CLAHE (Contrast Limited Adaptive Histogram Equalization).
• Introduced RandomBrightnessContrast and RandomGamma augmentations with a disabled state (probability set to 0.0).
• Included ImageCompression with a default quality setting of 75 but disabled it (probability set to 0.0).

🎯 Purpose & Impact

• 🎨 Enhanced Data Augmentation: Adding new transformations like CLAHE enriches the data augmentation pipeline and can potentially improve model robustness to various lighting conditions.
• 🔍 Fine-Tuning Probabilities: Adjusting the probabilities of existing augmentations aims to create a more balanced augmentation strategy that may lead to better generalization and prevent overfitting.
• 🚀 Experimentation Ready: Including new augmentations with a zero probability allows for quick experimentation. Developers can easily enable these augmentations in the future for further research and testing.
• 📈 Potential Impact: These changes could lead to more effective training and possibly improved accuracy without changing the current model behavior since new augmentations are initially set to not apply.