DESCRIPTION:

| Original image source: Flickr - Nicole Lee

⚠️ Important Note

• Input images are resized directly to fit the model’s expected input dimensions.
• Padding is not applied, which may reduce accuracy if your images have a different aspect ratio than the model’s input size.
• To maintain performance, try to use input sizes that closely match the aspect ratio of your images and try to reduce the size of the Image if possible.

---

📦 Requirements for the Program

• Refer to the requirements.txt file for dependencies.
• If you're using a NVIDIA GPU, install onnxruntime-gpu. Otherwise, install the standard onnxruntime.

---

🛠️ Installation

git clone https://github.com/ibaiGorordo/ONNX-YOLOv8-Object-Detection.git
cd ONNX-YOLOv8-Object-Detection
pip install -r requirements.txt

ONNX Runtime Setup

• For systems with NVIDIA GPU:

  pip install onnxruntime-gpu

• For CPU-only systems(without Discrete GPU):

  pip install onnxruntime

---

🔄 Convert YOLOv8 to ONNX(Open Neural Network Exchange)

Use the Google Colab notebook to export the model:

Alternatively, convert using Python with Ultralytics:

from ultralytics import YOLO

model = YOLO("yolov8m.pt")
model.export(format="onnx", imgsz=[480, 640])

---

💡 The original YOLOv8 models were converted to ONNX and other formats by PINTO0309. You can download them from his model zoo repository. Either run the download_single_batch.sh script or manually download and place the ONNX files (e.g., yolov8m_480x640.onnx) into the models directory. Make sure to update your script with the correct filename if needed.

---

📁 Original YOLOv8 Repository

• Source: ultralytics/ultralytics
• License: GPL-3.0 License

---

🚀 Example Usage:

🔹 Image Inference Used

python image_object_detection.py

🔹 Webcam Inference

python webcam_object_detection.py

---

🔗 References

• Ultralytics YOLOv8
• PINTO0309 Model Zoo
• Model Conversion Tools (OpenVINO → TensorFlow)

---