feature/add_tf_and_tflite_export_2

[SkalskiP]

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Adding explicit TFLite model support and TensorFlow export functionality to Ultralytics' YOLOv5.

📊 Key Changes

• Added TFLite and TensorFlow *.pb model extension support in .gitignore.
• Provided an option to disable layer fusion during model loading in attempt_load function for better compatibility with TensorFlow's layer behavior.
• Implemented an extensive new tf.py script to facilitate TensorFlow and TFLite model conversion, support, and inference operations for YOLOv5 models.
• Augmented the requirements.txt with a specific TensorFlow version required for TFLite export (tensorflow==2.4.1).

🎯 Purpose & Impact

• 🚀 Enabling broader adoption: With the ability to convert YOLOv5 models to TensorFlow and TFLite formats, developers can now deploy YOLOv5 on a broader range of platforms and devices, including those supporting TensorFlow Lite.
• 🔌 Greater flexibility: Users can now choose to load models with or without layer fusion, providing more control based on different deployment needs.
• 🧠 Facilitating quantization: TFLite export supports FP16 and Int8 quantization, thereby enabling efficient deployment on edge devices with limited computational resources.
• 🤖 Broader inference possibilities: Detecting objects can now be performed using TensorFlow models, leveraging its vast ecosystem.
• 🏹 Streamlined workflow: By integrating conversion scripts and compatibility changes directly into YOLOv5's repository, the process of taking a model from training to deployment on TensorFlow-compatible environments is greatly simplified.

[glenn-jocher]

@SkalskiP let's close this PR and delete this branch as #4479 has taken over this task now. Thanks!