Skip to content
/ GPU-Median-Blurring Public

This project enhances median blurring for 'salt and pepper' noise removal in images/videos. Three GPU acceleration methods (OpenACC, CUDA Python, CUDA C/C++) optimize the process. Challenges include real-time processing of high-resolution content. Findings: OpenACC (600x speedup), CUDA Python (1780x), CUDA C/C++ (2700x).

License

View license
2 stars 0 forks Branches Tags Activity
Star
Notifications You must be signed in to change notification settings

Alawaji214/GPU-Median-Blurring

BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

100 Commits
c
c
 
 
python
python
 
 
resources
resources
 
 
.gitignore
.gitignore
 
 
LICENSE.md
LICENSE.md
 
 
README.md
README.md
 
 
run_all_cuda.sh
run_all_cuda.sh
 
 

Repository files navigation

GPU Median Blurring

Building

Requirements

  • OpenCV library
  • Python and pip
  • Nvidia GPU ( this readme assumses Nvidia L4)

Additional Requirements

For building and running CUDA and OpenACC versions, ensure the following are installed:

Note: Please make sure to add all dependencies to your PATH.

OpenCV Installation

install opencv library

sudo apt install libopencv-dev

Pything Libraries

for python, you have to install opencv through pip

pip3 install opencv-python
pip3 install numba

C

OpenACC

To compile and run the OpenACC version:

Navigate to the OpenACC directory:

cd c/openacc

compile the application

nvc++ -acc -gpu=cc89 -Minfo=acc -I/usr/include/opencv4 -L/usr/lib -lopencv_core -lopencv_imgproc -lopencv_imgcodecs -lopencv_highgui median_filter.cpp -o median_filter_gpu
./median_filter_gpu

Or, for passing the image name as an argument:

nvc++ -acc -gpu=cc89 -Minfo=acc -I/usr/include/opencv4 -L/usr/lib -lopencv_core -lopencv_imgproc -lopencv_imgcodecs -lopencv_highgui median_filter_args.cpp -o median_filter_gpu_args
./median_filter_gpu_args <imageName>

To profile the OpenACC application:

Copy code
nsys profile -t openacc -b dwarf --stats=true ./median_filter

Cuda

goto c directary and use cmake to build the application

cd c
mkdir build && cd build
cmake ..
make

and then if everything run smothly, you can launch the application

./median #seq c
./image_cuda_median #cuda c
./video_cuda_median #cuda c
./video_cuda_median videos_2_1080p.mp4

or alternatively, you can use the following link to dirtectly build it

g++ -std=c++11 -I/usr/local/include/opencv4  -L/usr/local/lib -lopencv_core -lopencv_imgproc -lopencv_imgcodecs -lopencv_highgui median_filter.cpp -o median_filter

Python

nosiy image: noisy_image1.jpg

cd python
pwd
/opt/nvidia/hpc_sdk/Linux_x86_64/23.11/compilers/bin/nsys profile --stats=true --output /home/bandr1994/COE506-Project/python/median_filter python median_filter.py ../resources/noisy_image1.jpg
/opt/nvidia/hpc_sdk/Linux_x86_64/23.11/compilers/bin/nsys profile --stats=true --output /home/bandr1994/COE506-Project/python/median_filter python median_filter_jit.py ../resources/noisy_image1.jpg
/opt/nvidia/hpc_sdk/Linux_x86_64/23.11/compilers/bin/nsys profile --stats=true python median_filter_multi.py ../resources/noisy_image1.jpg
/opt/nvidia/hpc_sdk/Linux_x86_64/23.11/compilers/bin/nsys profile --stats=true python median_filter_opencv.py ../resources/noisy_image1.jpg
/opt/nvidia/hpc_sdk/Linux_x86_64/23.11/compilers/bin/nsys profile --stats=true --output /home/bandr1994/COE506-Project/python/median_filter_cuda python median_filter_cuda.py ../resources/noisy_image1.jpg
/opt/nvidia/hpc_sdk/Linux_x86_64/23.11/compilers/bin/nsys profile --stats=true python median_filter_cuda_multi.py ../resources/noisy_image1.jpg
/opt/nvidia/hpc_sdk/Linux_x86_64/23.11/compilers/bin/nsys profile --stats=true python median_filter_cuda_opt.py ../resources/noisy_image1.jpg


/opt/nvidia/hpc_sdk/Linux_x86_64/23.11/compilers/bin/nsys profile --stats=true python meidan_filter_vedio.py ../resources/video/videos_1_1080p.mp4
/opt/nvidia/hpc_sdk/Linux_x86_64/23.11/compilers/bin/nsys profile --stats=true python meidan_filter_vedio_cuda.py ../resources/video/videos_1_1080p.mp4

/opt/nvidia/hpc_sdk/Linux_x86_64/23.11/compilers/bin/nsys profile --stats=true --output /home/bandr1994/COE506-Project/python/median_filter_cuda python median_filter_cuda.py ../resources/noisy_image2.jpg

/opt/nvidia/hpc_sdk/Linux_x86_64/23.11/compilers/bin/nsys profile --stats=true --output /home/bandr1994/COE506-Project/python/median_filter_cuda python median_filter_cuda.py ../resources/noisy_image3.jpg

/opt/nvidia/hpc_sdk/Linux_x86_64/23.11/compilers/bin/nsys profile --stats=true --output /home/bandr1994/COE506-Project/python/median_filter_cuda python median_filter_cuda.py ../resources/noisy_image4.jpg

/opt/nvidia/hpc_sdk/Linux_x86_64/23.11/compilers/bin/nsys profile --stats=true --output /home/bandr1994/COE506-Project/python/median_filter_cuda python median_filter_cuda.py ../resources/noisy_image5.jpg

Results

Used GPU

NVIDIA L4

output of nvidia-smi

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 525.105.17   Driver Version: 525.105.17   CUDA Version: 12.0     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA L4           On   | 00000000:00:03.0 Off |                    0 |
| N/A   47C    P8    17W /  72W |      0MiB / 23034MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
                                                                   
+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

output of /usr/local/cuda-11.8/extras/demo_suite/deviceQuery

/usr/local/cuda-11.8/extras/demo_suite/deviceQuery Starting...

 CUDA Device Query (Runtime API) version (CUDART static linking)

Detected 1 CUDA Capable device(s)

Device 0: "NVIDIA L4"
  CUDA Driver Version / Runtime Version          12.0 / 11.8
  CUDA Capability Major/Minor version number:    8.9
  Total amount of global memory:                 22518 MBytes (23612293120 bytes)
MapSMtoCores for SM 8.9 is undefined.  Default to use 128 Cores/SM
MapSMtoCores for SM 8.9 is undefined.  Default to use 128 Cores/SM
  (58) Multiprocessors, (128) CUDA Cores/MP:     7424 CUDA Cores
  GPU Max Clock rate:                            2040 MHz (2.04 GHz)
  Memory Clock rate:                             6251 Mhz
  Memory Bus Width:                              192-bit
  L2 Cache Size:                                 50331648 bytes
  Maximum Texture Dimension Size (x,y,z)         1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)
  Maximum Layered 1D Texture Size, (num) layers  1D=(32768), 2048 layers
  Maximum Layered 2D Texture Size, (num) layers  2D=(32768, 32768), 2048 layers
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       49152 bytes
  Total number of registers available per block: 65536
  Warp size:                                     32
  Maximum number of threads per multiprocessor:  1536
  Maximum number of threads per block:           1024
  Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
  Max dimension size of a grid size    (x,y,z): (2147483647, 65535, 65535)
  Maximum memory pitch:                          2147483647 bytes
  Texture alignment:                             512 bytes
  Concurrent copy and kernel execution:          Yes with 2 copy engine(s)
  Run time limit on kernels:                     No
  Integrated GPU sharing Host Memory:            No
  Support host page-locked memory mapping:       Yes
  Alignment requirement for Surfaces:            Yes
  Device has ECC support:                        Enabled
  Device supports Unified Addressing (UVA):      Yes
  Device supports Compute Preemption:            Yes
  Supports Cooperative Kernel Launch:            Yes
  Supports MultiDevice Co-op Kernel Launch:      Yes
  Device PCI Domain ID / Bus ID / location ID:   0 / 0 / 3
  Compute Mode:
     < Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 12.0, CUDA Runtime Version = 11.8, NumDevs = 1, Device0 = NVIDIA L4
Result = PASS

to profile nsys

nsys profile --stats=true ./cuda_median

noisy_image1

Method Total Time Kernal Time GPU Transfer Time
OpenCV
Seq C 153 sec
CUDA C 0.053 sec 0.046 sec 0.007 sec
OpenACC 0.256 sec 0.231 sec 0.026 sec
CUDA Pythn 0.086 sec 0.060 sec 0.026 sec

noisy_image2

Method Total Time Kernal Time GPU Transfer Time
OpenCV
Seq C 152 sec
CUDA C 0.06409 sec 0.05772 sec 0.00637 sec
OpenACC 0.26040 sec 0.23297 sec 0.02743 sec
CUDA Pythn 0.08495 sec 0.05970 sec 0.02525 sec

noisy_image3

Method Total Time Kernal Time GPU Transfer Time
OpenCV
Seq C 250 sec
CUDA C 0.10428 sec 0.09386 sec 0.01042 sec
OpenACC 0.42392 sec 0.37996 sec 0.04396 sec
CUDA Pythn 0.13956 sec 0.09770 sec 0.04186 sec

noisy_image4

Method Total Time Kernal Time GPU Transfer Time
OpenCV
Seq C 0.344 sec
CUDA C 0.00019 sec 0.00017 sec 0.00002 sec
OpenACC 0.00067 sec 0.00051 sec 0.00015 sec
CUDA Pythn 0.00022 sec 0.00020 sec 0.00002 sec

noisy_image5

Method Total Time Kernal Time GPU Transfer Time
OpenCV
Seq C 0.58 sec
CUDA C 0.00035 sec 0.00032 sec 0.00003 sec
OpenACC 0.00111 sec 0.00088 sec 0.00023 sec
CUDA Pythn 0.00028 sec 0.00025 sec 0.00003 sec

sp_img_gray_noise_heavy

Method Total Time Kernal Time GPU Transfer Time
OpenCV
Seq C 0.73 sec
CUDA C 0.118 sec 0.001 sec 0.116 sec
OpenACC 1.367 ms 1.171 ms 0.042 ms
CUDA Pythn 0.429 ms 0.208 ms 0.021 ms

License

This project is licensed under the terms of the custom license. See the LICENSE file for details.

About

This project enhances median blurring for 'salt and pepper' noise removal in images/videos. Three GPU acceleration methods (OpenACC, CUDA Python, CUDA C/C++) optimize the process. Challenges include real-time processing of high-resolution content. Findings: OpenACC (600x speedup), CUDA Python (1780x), CUDA C/C++ (2700x).

Resources

Readme

License

View license
Activity

Stars

2 stars

Watchers

1 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •  

Languages