CUDA Image Processing

A high-performance GPU-accelerated image processing application implemented in C++ with CUDA. This project aims to achieve at least 70% throughput improvement over CPU-only implementations.

Features

• Gaussian Smoothing: Separable convolution implementation for efficient memory usage
• Edge Detection: Sobel and Canny edge detection with shared memory optimization
• Histogram Equalization: GPU-accelerated histogram computation and equalization
• Multi-format Support: Both grayscale and RGB image processing
• Performance Benchmarking: Comprehensive CPU vs GPU performance comparison

Technical Implementation

• CUDA C++ kernels for GPU computation
• OpenCV for image I/O operations
• CUDA streams for overlapping data transfers and computation
• Optimized memory coalescing and shared memory usage
• Automatic kernel launch parameter tuning

Build Requirements

• CMake 3.18+
• CUDA Toolkit 11.0+
• OpenCV 4.0+
• C++17 compatible compiler
• GPU with compute capability 7.5+ (RTX 20xx series or newer recommended)

Building

mkdir build
cd build
cmake .. -DCMAKE_BUILD_TYPE=Release
make -j

Usage

./cuda-image-proc --mode gpu --ops gaussian,sobel --batch input_dir/

Performance Goals

Target: ≥70% throughput improvement over CPU implementations on 1080p image batches.

Project Structure

├── src/           # Source code
├── tests/         # Unit tests
├── cmake/         # CMake modules
└── docs/          # Documentation