soft953 / gpu_erosion_dilatation

GPGPU_Project: Implement CUDA version of an image processing algorithm

💻 CUDA, C++, CMake

• CPU and GPU (Cuda) implementation of binary opening/closing

• Library Needed: Opencv

• Run CPU: src/ Cmake ./gpgpu path_src

• Run GPU: cuda/ Makefile ./main path_src name_dst

• Test images: build/

Result: (1000x1000) image with 5x5 kernel

Benchmark:

Kernel size: 5x5 - i7-8750H 6 coeurs/12 threads - Fréquence 2,20 GHz - 4,10 GHz et un GPU Nvidia 1050Ti 4Go 768 CUDA Cores

Center-aligned	Center-aligned	Center-aligned	Center-aligned
128x128	32 ms	0.06 ms	0.056 ms
256x256	37 ms	0.099 ms	0.1 ms
600x600	83 ms	0.359 ms	0.386 ms
800x800	140 ms	1.074 ms	0.989 ms
1000x1000	184 ms	1.383 ms	1.11 ms
4k	1278 ms	32.705 ms	17.26 ms
8k	4848 ms	123.230 ms	62.111 ms