Project

Edge Detection Filter

Introduction

In this project, we implemented an Edge Detection Filter based on Sobel Filter. The project’s core is based on SIMD -or, to be more precise, SIMT- computations using NVIDIA CUDA. The project features both CLI and GUI. The GUI is written in Python, uses the Tkinter library, and communicates with the CUDA core, which is written in C++ using a shell.

For more information regarding the implementation and analysis of the result, read Document.pdf.

Compilation

To be able to compile the core, you should have previously installed OpenCV and NVIDIA Cuda compiler.

To compile the core, run the following command:

nvcc main.cu `pkg-config opencv --cflags --libs` -o main.out

Note

Based on the version of OpenCV and the system you are using, you may need to change the command like this:

nvcc main.cu -arch=sm_86 `pkg-config opencv4 --cflags --libs` -o main.out

Also, note that sm_86 is used for a GPU with 8.6 compute capability. You should change it based on your GPU Compute Capability. You can find your GPU Compute Capability using the following command:

nvidia-smi --query-gpu=compute_cap --format=csv,noheader|head -n 1

Run

GUI

Put the main.out file in the same directory as main.py and run the following command:

python GUI/main.py

CLI

Run main.out with the following arguments:

./main.out [image_path] [alpha] [beta] [thresh]

• alpha: Desired Contrast Value
• beta: Desired Brigtness Value
• thresh: Desired Sobel Filter Threshold

Demo