https://github.com/stefanolusardi/resnet-onnx
ResNet & SqueezeNet inference built using ONNX Runtime C++, without OpenCV
https://github.com/stefanolusardi/resnet-onnx
cmake computer-vision cpp onnx onnxruntime resnet squeezenet
Last synced: 7 months ago
JSON representation
ResNet & SqueezeNet inference built using ONNX Runtime C++, without OpenCV
- Host: GitHub
- URL: https://github.com/stefanolusardi/resnet-onnx
- Owner: StefanoLusardi
- Created: 2025-01-21T17:12:05.000Z (9 months ago)
- Default Branch: main
- Last Pushed: 2025-01-21T22:18:10.000Z (9 months ago)
- Last Synced: 2025-01-21T22:28:35.810Z (9 months ago)
- Topics: cmake, computer-vision, cpp, onnx, onnxruntime, resnet, squeezenet
- Language: C++
- Homepage:
- Size: 1.82 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# ResNet-ONNX
Welcome to ResNet-ONNX!
This is a project to show how to perform inference on a ResNet or SqueezeNet models using C++.
In order to simplify dependencies management as well as to improve performance, OpenCV is not required.
The algorithms contained in [pre_processing.hpp](./src/pre_processing.hpp/) and [post_processing.hpp](./src/post_processing.hpp/) are written in plain C++.
The only dependency is [stb_image.h](https://github.com/nothings/stb/blob/master/stb_image.h) which is used to load images from disk.
## Getting Started
This project uses git submodules so it is required to clone it using the *--recurse-submodules* flag in order to retrive the required submodules.
```bash
# Clone the repo
git clone https://github.com/StefanoLusardi/ResNet-ONNX --recurse-submodules
```
In order to kickstart this project it is required to configure a local development environment using the provided [scripts](./scripts/).
The only pre-requisite is to have a recent (i.e. >= 3.8) Python installed on your machine.
The development environment uses python-virtualenv to create a local environment in which all the project dependencies are installed (exactly as virtual environment is supposed to work for any ordinary Python projects). The virtual environment will be created in the *.venv* folder in the root of the repo.
Third party C++ dependencies will be installed using Conan Package Manager in a local cache folder. The conan cache will be created in the *.conan* folder in the root of the repo.
These preliminary steps must be done only the first time the project is bootstrapped: afterwards it is only required to activate the virtual environment in order to work with the project itself.
## Create development environment
In order to setup a development environment it is sufficient to run the script *scripts/env/setup.* depending on your operating system.
```bash
# Linux/MacOS
chmod +x scripts/env/setup.sh
scripts/env/setup.sh
# Windows
scripts\env\setup.bat
```
## Install Python packages
Run this command into the previously created python virtual environment in order to install the required packages to download, export and validate the Yolo model.
```bash
pip install -r models/requirements.txt
```
## Download ResNet and SqueezeNet models
Run these commands to download a pre-trained ResNet and SqueezeNet models from the ONNX Model Zoo repository.
```bash
python models/download_models.py
```
After these commands you should have *resnet18-v2-7.onnx* and *squeezenet1.1-7.onnx* binaries into the root of your repository. In case you want to retrieve other versions of the binaries it is enough to replace the url into the *models/download_models.py* script.
## Start development environment
In order to start the development environment it is sufficient to activate the Python Virtual Environment just created the step above.
```bash
# Linux/MacOS
source .venv/bin/activate
# Windows
.venv\Scripts\activate.bat
```
## Setup Build Environment (Windows Only)
When building from command line on Windows it is necessary to activate the Visual Studio Developer Command Prompt.
Depending on the version of Visual Studio compiler and on its install location it is required to run *vcvars64.bat* script the set the development environment properly.
*Note*: using Visual Studio IDE or the CMake extension for VSCode this step is already managed in the background, so no action is required.
Examples:
```bash
# Visual Studio 2022 - Build Tools
"C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Auxiliary\Build\vcvars64.bat"
# Visual Studio 2019 - Enterprise
"C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat"
```
## Setup Dependencies
Install the project dependencies as specified in the conanfile.txt.
```bash
python scripts/conan/setup.py
# examples:
python scripts/conan/setup.py Release clang 15
python scripts/conan/setup.py Debug visual_studio 17
# Visual Studio versions:
# VS 2019: 16
# VS 2022: 17
```
## Build
Run the following script to configure, build and install the project.
```bash
python scripts/cmake.py
# examples:
python scripts/cmake.py Release clang 15
python scripts/cmake.py Debug visual_studio 17
```
## Run
Once the project has been installed you can run it using the following command.
```bash
install/resnet
```
This is an example of the expected output using SqueezeNet:
```bash
Model: squeezenet1.1-7.onnx
Image: "images/dog.jpeg"
Input: 0
- name: data
- shape: 1x3x224x224
- element type: 1
Output: 0
- name: squeezenet0_flatten0_reshape0
- shape: 1x1000
- element type: 1
Prediction results:
ID: 232
Label: Border collie
Confidence: 0.886919
```
This is an example of the expected output using ResNet:
```bash
Model: resnet18-v2-7.onnx
Image: "images/dog.jpeg"
Input: 0
- name: data
- shape: 1x3x224x224
- element type: 1
Output: 0
- name: resnetv22_dense0_fwd
- shape: 1x1000
- element type: 1
Prediction results:
ID: 232
Label: Border collie
Confidence: 0.871716
```