2024 NYCU PP Final Project

Authors: Ching-Yang Lin, Ying-Huan Chen, Chung-Ho Wu

CUDA-ISP (Image Signal Processing) is a high-performance image processing pipeline designed for CUDA-enabled GPUs. This repository provides a flexible framework for performing various image processing tasks, such as degradation, CPU-based processing, and GPU-accelerated processing.

• Degradation Simulation: Generate degraded images from input PNG files.
• CPU-Based Image Processing: Reference implementation of the processing pipeline on the CPU.
• GPU-Accelerated Image Processing: High-performance implementation of the pipeline using CUDA.

• CUDA-enabled GPU
• NVIDIA CUDA Toolkit
• A C++ compiler supporting C++11
• Linux-based operating system (tested on Ubuntu)

1. Clean any existing build artifacts:

   make clean

2. Compile the project:

   make

The resulting executable will be named cuda-isp.

./cuda-isp <input_image> <output_image> <mode>

• <input_image>: Path to the input image.
• <output_image>: Path to the output image.
• <mode>: Processing mode:
  • degrad: Degradation simulation.
  • cpu: CPU-based processing.
  • gpu: GPU-accelerated processing.

The ex_single.sh script demonstrates how to use the tool for various image sizes:

make clean
make
mkdir -p output/
./cuda-isp images/test_3000x2000.png images/test_raw_3000x2000.bmp degrad
./cuda-isp images/test_raw_3000x2000.bmp output/output_cpu_3000x2000.bmp cpu
./cuda-isp images/test_raw_3000x2000.bmp output/output_gpu_3000x2000.bmp gpu
./cuda-isp images/test_6000x4000.png images/test_raw_6000x4000.bmp degrad
./cuda-isp images/test_raw_6000x4000.bmp output/output_cpu_6000x4000.bmp cpu
./cuda-isp images/test_raw_6000x4000.bmp output/output_gpu_6000x4000.bmp gpu
./cuda-isp images/test_12000x8000.png images/test_raw_12000x8000.bmp degrad
./cuda-isp images/test_raw_12000x8000.bmp output/output_cpu_12000x8000.bmp cpu
./cuda-isp images/test_raw_12000x8000.bmp output/output_gpu_12000x8000.bmp gpu

.
├── filters/         # Custom image processing filters
├── stb_image/       # Dependency for image loading and saving
├── images/          # Example input images
├── output/          # Generated output images
├── main.cu          # Main CUDA implementation
├── Makefile         # Build script
├── ex_single.sh     # Example usage script
└── README.md        # Project documentation

This framework leverages CUDA to accelerate image processing tasks, achieving significant speed-ups compared to CPU-based processing. GPU implementations are particularly effective for high-resolution images (e.g., 6000x4000 or larger).