Feature/classification tutorial refactor

[paulguerrie]

Refactored the classification tutorial to match the main tutorial.

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Added a comprehensive tutorial notebook for YOLOv5 classification tasks.

📊 Key Changes

• 📁 New tutorial.ipynb notebook added for classification.
• 🚀 Includes setup, prediction, validation, and training guidelines.
• ✨ Features tensorboard integration for logging and visualization.
• 🎓 Instructions for active learning loop closure using Roboflow.
• 🧪 Comet and ClearML integration details provided.
• 🔒 Local logging with CSV and TensorBoard explained.
• 📚 Additional appendix content for further learning and execution.

🎯 Purpose & Impact

• 📖 Provides an end-to-end tutorial for users interested in YOLOv5 classification to follow along easily.
• 🛠️ Facilitates easier setup and use of YOLOv5's classification capabilities, potentially increasing adoption and user success.
• 🏋️‍♂️ Helps users understand and apply transfer learning for their custom classification tasks.
• 🌐 Enhances user experience by integrating with robust experiment tracking and dataset versioning systems.

[paulguerrie]

Here's the predict output image for you guys to host.

[glenn-jocher]

@paulguerrie thanks! I'll take a look. Did you run inference at --img 224 for zidane.jpg? Confidence seem pretty low.

[paulguerrie]

I don't think so. I'm pretty sure the image I provided was the output of this command: !python classify/predict.py --weights yolov5s-cls.pt --img 640 --source data/images

[glenn-jocher]

I don't think so. I'm pretty sure the image I provided was the output of this command: !python classify/predict.py --weights yolov5s-cls.pt --img 640 --source data/images

Ohh ok got it. That makes sense then. The ImageNet-trained classification models are all at --imgsz 224, so running them at 640 will cause poor results. Detection and Segmentation are all at 640 though :)

[glenn-jocher]

@paulguerrie hey I'm updating all notebooks now, just noticed there seems to be an image missing here. I think I'm going to merge this now, but if you can provide me the updated image URL I can fix this. This should probably be an image of Roboflow classification annotation, or just annotation in general.

[glenn-jocher]

EDIT, I'll just use this one from the blog post :)

[glenn-jocher]

@paulguerrie PR is merged. Thank you for your contributions to YOLOv5 🚀 and Vision AI ⭐