https://github.com/myzhar/opencv_cpu_vs_gpu

A serie of tests to compare performances of CPU and GPU processing
https://github.com/myzhar/opencv_cpu_vs_gpu

Last synced: 8 days ago
JSON representation

A serie of tests to compare performances of CPU and GPU processing

• Host: GitHub
• URL: https://github.com/myzhar/opencv_cpu_vs_gpu
• Owner: Myzhar
• License: gpl-3.0
• Created: 2017-03-26T07:01:50.000Z (almost 8 years ago)
• Default Branch: master
• Last Pushed: 2017-04-26T07:49:17.000Z (over 7 years ago)
• Last Synced: 2024-11-08T13:15:51.476Z (about 2 months ago)
• Language: C++
• Size: 3.88 MB
• Stars: 3
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# opencv_cpu_vs_gpu

A serie of tests to compare performances of **CPU** and **GPU** processing.

This benchmark is based on OpenCV 2.4.13 and performs a simple basic algorithm of computer vision :

* **resize**: to keep costant the process time and indipendent by image size

* **color conversion**: to pass from RGB to grayscale image

* **blur**: to remove noises

* **Canny**: to detect image corners

The algorithm has not a well defined goal, but it may be for example the beginning of a process of "line" or "circle" detection...

The process is iterated "N" times on the same image and a mean of the single step times and of the total time is calculated.

The same algorithm is moved to the **GPU** (only if the machine is *CUDA enabled*).

**GPU TEST**

* *Classic memory copy*: uses the "upload" approach to copy memory from host to device

* *ZERO COPY*: uses "gpu::CudaMem" with "ALLOC_ZEROCOPY" flag to take advantage of shared memory *(if available)*

* *Memory Managed*: allocates memory using "cudaMallocManaged" to take advantage of pinned memory *(if available)*