Create YOLOv5 BaseModel class

[glenn-jocher]

Creates 4 new model classes.

• BaseModel(nn.module)
• DetectionModel(BaseModel) <-- equivalent to current Model(nn.module)
• ClassificationModel(BaseModel)
• SegmentationModel(BaseModel) <-- available in theory, not created in this PR

Defines DetectionModel = Model for backwards compatibility, where current YOLOv5 models are of type models.yolo.Model

@AyushExel please review

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Refactor classifier script and introduce YOLOv5 ClassificationModel.

📊 Key Changes

• Created a ClassificationModel class, derived from the new BaseModel class structure.
• Removed redundant code and cleaned up the importing of modules and methods.
• smart_hub_load method introduced to handle model loading more gracefully, especially considering the new Torch 1.12 version requirements.
• Simplified data loader functionality by always using InfiniteDataLoader instead of choosing between it and DataLoader.

🎯 Purpose & Impact

• 🔨 Refactoring: The changes aim to streamline classifiers by creating a more structured and distinguished classification model, simplifying the codebase.
• 🆕 Feature Introduction: The new smart_hub_load method is introduced to handle model downloads and loading more efficiently, reducing potential errors and improving compatibility with latest PyTorch versions.
• 🚀 Performance: Cleaner and more maintainable code often leads to fewer bugs and easier future enhancements.
• ⏱️ Efficiency: By always using InfiniteDataLoader, the data loading process is potentially more consistent and easier to manage.
• 💡 User Experience: These changes are set to improve the overall experience for developers using the Ultralytics YOLOv5 classifier, providing a solid and simpler foundation for building classification models.

[AyushExel]

Looks good! Please merge
Great idea. we should keep the same naming convention for all tasks? Or maybe use yoloDetector, yoloClassifier, yoloPoseEstimator .. etc? I prefer what you've got here.

[glenn-jocher]

@AyushExel awesome, its done!