While running inference using YoloV3 onnx model, I got the following error :-

ValueError: This ORT build has ['TensorrtExecutionProvider', 'CUDAExecutionProvider', 'CPUExecutionProvider'] enabled. Since ORT 1.9, you are required to explicitly set the providers parameter when instantiating InferenceSession. For example, onnxruntime.InferenceSession(..., providers=['TensorrtExecutionProvider', 'CUDAExecutionProvider', 'CPUExecutionProvider'], ...)

On checking the code, in /models/common.py , after line 315, cuda = torch.cuda.is_available() was missing and also providers for onnx was missing. Hence by referring to Onnx inference code in YoloV5 (Yolov5/models/common.py), I have added the following lines of code

cuda = torch.cuda.is_available()
check_requirements(('onnx', 'onnxruntime-gpu' if cuda else 'onnxruntime'))
import onnxruntime
providers = ['CUDAExecutionProvider', 'CPUExecutionProvider'] if cuda else ['CPUExecutionProvider']
session = onnxruntime.InferenceSession(w, providers=providers)

Have tested the code with CPU and GPU both and inferencing is running now.

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Improvement in ONNX Runtime inference for the YOLOv3 model by selectively enabling GPU support.

📊 Key Changes

• Detected whether CUDA (GPU support) is available and set the variable cuda accordingly.
• Modified requirement checks to use onnxruntime-gpu when CUDA is available, otherwise defaulting to onnxruntime.
• Specified ONNX Runtime inference providers to include GPU support if CUDA is present.

🎯 Purpose & Impact

• Increased Flexibility: By automatically detecting GPU availability, the model can now leverage accelerated computation when possible, leading to performance improvements.
• Smoother User Experience: Users no longer need to manually specify whether they're using a GPU; the system intelligently adapts, simplifying the setup process.
• Potential Impact: Improved performance and usability can enhance the applicability of YOLOv3 across various environments, benefiting those requiring efficient real-time object detection.