New DetectMultiBackend() class

[glenn-jocher]

Consolidates all backends into a single YOLOv5 DetectMultiBackend() class. Supports following model formats:

• PyTorch *.pt
• TorchScript *.torchscript.pt
• CoreML *.mlmodel
• TensorFlow saved_model
• TensorFlow *.pb
• TensorFlow Lite *.tflite
• TensorFlow.js web_model
• ONNX Runtime *.onnx
• OpenCV DNN *.onnx

Usage

# Export
python export.py --weights yolov5s.pt --include tflite

# Inference
python detect.py --weights yolov5s.tflite
python val.py --weights yolov5s.tflite

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Refactoring and optimization of YOLOv5 inference and export.

📊 Key Changes

• 📥 Modified the download URL in the coco128.yaml file.
• 🧹 Code cleanup in detect.py by removing unused imports and refactoring model loading code.
• 🤖 Introduced DetectMultiBackend class to models/common.py for flexible backend support during inference.
• 🔍 Added JSON export option with model metadata in PyTorch export in export.py.
• 🤏 Half-precision (FP16) support and device selection improvements for inference code.
• 👁‍🗨 Enhanced DetectMultiBackend to include visualization features.
• 🔗 Major refactoring in detect.py, val.py, and export.py to utilize new backend class and streamline the codebase.
• ❌ Removed load_classifier function from utils/torch_utils.py as it was deprecated.

🎯 Purpose & Impact

• 🚀 The changes aim to improve code clarity, maintainability, and execution efficiency.
• ✨ These updates lay the groundwork for more versatile model deployment by unifying the inference pipeline across different computing backends.
• 👤 Users will benefit from a more streamlined inference process, potentially faster execution times, and easier integration with various deployment environments.

[glenn-jocher]

Ok this is PR is ready to merge!

@zldrobit it doesn't make sense to include TensorFlow.js in here because this model needs a javascript environment to run right?

[zldrobit]

@glenn-jocher I agree with you. It doesn't make sense, since TesnorFlow.js requires a browser or node.js environment to run and can only be called by javascript.