LibtorchOpenCVTutorials

A comprehensive C++ project demonstrating machine learning and deep learning concepts using LibTorch (PyTorch C++ API), OpenCV, and related libraries. Includes implementations of Linear Regression, Logistic Regression, Neural Networks (ANN, CNN), Autograd, Tensor operations, and more.

Version

• LibTorch: 2.9.1 (CUDA 12.6)
• OpenCV: 4.x (installed via vcpkg)
• C++ Standard: C++20
• CMake: 3.20+
• CUDA: 12.6

---

Prerequisites

Install the following dependencies before building the project.

1. Required System Packages

# Build tools
sudo apt update
sudo apt install build-essential cmake curl wget git pkg-config

# CUDA (must match LibTorch version - CUDA 12.6)
# Install NVIDIA CUDA Toolkit 12.6 from: https://developer.nvidia.com/cuda-downloads

# Required system libraries
sudo apt install libpthread-stubs0-dev libcurl4-openssl-dev

2. vcpkg (Package Manager)

git clone https://github.com/microsoft/vcpkg.git ~/vcpkg
~/vcpkg/bootstrap-vcpkg.sh

3. Install vcpkg Packages

The project requires the following vcpkg packages (LibTorch is NOT installed via vcpkg - see next section):

cd ~/vcpkg

# Install required packages (listed individually for clarity)
./vcpkg install opencv4

./vcpkg install benchmark
./vcpkg install curl
./vcpkg install pthread

Note: The csv-parser and smartStore libraries are included as git submodules in the external/ directory and do not require separate installation.

4. Download & Link LibTorch (2.9.1 CUDA 12.6)

LibTorch is not installed via vcpkg. You must download and place it in the external/libtorch/ directory manually.

# Navigate to the project root
cd /path/to/LibtorchOpenCVTutorials

# Create external directory if it doesn't exist
mkdir -p external

# Download LibTorch C++ distribution (CUDA 12.6 version)
wget https://download.pytorch.org/libtorch/cu126/libtorch-cxx11-abi-shared-with-deps-2.9.1%2Bcu126.zip

# Unzip directly into external/
unzip libtorch-cxx11-abi-shared-with-deps-2.9.1+cu126.zip -d external/

# Verify structure
ls external/libtorch/lib/libtorch.so   # should exist

Directory structure after download:

external/
  libtorch/
    include/         # LibTorch headers
    lib/             # LibTorch shared libraries (.so files)
    share/           # CMake config files

Important: The CMakeLists.txt links directly to the .so files in external/libtorch/lib/ using absolute paths. Ensure the directory structure matches the expected layout.

5. Configure vcpkg Toolchain in CMake

Ensure your CMake configuration uses the vcpkg toolchain file. Either set the CMAKE_TOOLCHAIN_FILE variable or use the -D flag during configuration.

Option A: Environment variable

export VCPKG_ROOT=~/vcpkg

Option B: CMake configuration with toolchain

cmake -B build -S . -DCMAKE_TOOLCHAIN_FILE=~/vcpkg/scripts/buildsystems/vcpkg.cmake

---

Building the Project

# From the project root
mkdir -p build
cd build

# Configure with vcpkg toolchain
cmake .. -DCMAKE_TOOLCHAIN_FILE=~/vcpkg/scripts/buildsystems/vcpkg.cmake

# Build
make -j$(nproc)

# Run the executable
./LibtorchOpenCVTutorials

---

Project Structure

├── CMakeLists.txt
├── main.cpp
├── include/
│   └── Util.h
├── databases/
│   ├── data-01-test-score.csv
│   ├── data-02-stock_daily.csv
│   ├── data-03-diabetes.csv
│   ├── data-04-zoo.csv
│   ├── fashion_minist_large.data
│   └── fmnist_small.csv
├── external/
│   ├── libtorch/                  # Manual download (NOT vcpkg)
│   ├── csv-parser/                # Git submodule
│   ├── smartStore/                # Git submodule
│   └── third_party/               # doctest
├── src/
│   ├── ANN/                       # ANN Fashion MNIST
│   ├── ANNGPUBASED/               # CNN & GPU-based implementations
│   ├── autogradsInLibtorch/       # Autograd examples
│   ├── BasicFunctionality/        # Tensor operations, Image reading
│   ├── LinearRegression/          # Single & Multi-variable Linear Regression
│   ├── LogisticRegression/        # Logistic Regression with data
│   ├── ReadingAndWriting/         # CSV reading, data validation
│   ├── TensorManipulation/        # Tensor manipulation utilities
│   └── TrainingPipeline/          # Neural Network training pipeline

---

MNIST Dataset Download

The MNIST dataset files are not included in the repository due to their large size. If the binary files are not already present in databases/mnist_data/, download and extract them manually:

# From the project root
mkdir -p databases/mnist_data/raw && cd databases/mnist_data/raw

# Download the 4 files
wget http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz
wget http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz
wget http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz
wget http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz

# Extract
gunzip *.gz

# Move files up one directory (out of raw/)
mv *.ubyte ..
cd ..
rmdir raw

After extraction, your databases/mnist_data/ folder should contain:

databases/mnist_data/
├── train-images-idx3-ubyte
├── train-labels-idx1-ubyte
├── t10k-images-idx3-ubyte
└── t10k-labels-idx1-ubyte

---

Key Features

• Linear Regression (Single & Multi-variable)
• Logistic Regression (Binary classification)
• Artificial Neural Networks (ANN for Fashion MNIST)
• Convolutional Neural Networks (CNN with GPU support)
• Autograd in C++ (automatic differentiation)
• Tensor operations & manipulation
• Data loading from CSV files
• Training pipeline with custom data loaders

---

License

This project is for educational purposes.