ResNet50v2 Image Classification with ONNX Runtime

This project demonstrates how to perform image classification using the ResNet50v2 model with ONNX Runtime in C++ and python programming languages. The program takes an input image, preprocesses it, runs inference using the model, and outputs the top predicted labels along with their probabilities.

Prerequisites

Before running the program, ensure you have the following dependencies installed:

• C++
  • C++ compiler supporting C++17 or later
  • CMake version 3.28 or later
  • OpenCV library (version 4.x)
  • ONNX Runtime C++ API
• Python
  • python 3.10 or later
  • Jupyter notebook

Building and running in C++

1.Clone this repository to your local machine:

git clone https://github.com/mbicl/onnx_inference.git

2.Build the program using CMake and run the compiled executable:

# building
mkdir build && cd build
cmake ../
make
#running
cd ..
./build/onnx_inference

3.Follow the on-screen instructions to enter the path to the input image.

4.View the top predicted labels and their probabilities printed on the console.

Configuration

• Model Path: By default, the program expects the ResNet50v2 ONNX model file (resnet50v2.onnx) to be located in the project directory. You can modify the model_path variable in the source code if the model file is located elsewhere.

• Label File: The program loads label names from a JSON file (imagenet-simple-labels.json). Ensure this file is present in the specified path. You can change the path by modifying the load_json function call in the source code.

Troubleshooting

• If you encounter any issues during compilation or execution, ensure that all dependencies are properly installed and paths are correctly specified.