University of Pennsylvania, CIS 565: GPU Programming and Architecture, Project 4
- Yash Vardhan
- Tested on: Windows 10 Pro, Intel i5-4200U (4) @ 2.600GHz 4GB, NVIDIA GeForce 840M 2048MB
In this project, I have implemented a image filters through the use of GPU and CPU and performed an analysis between them. I have implemented two awesome Computer Vision image filters like Bilateral Filter and HDR approximation filter(a.k.a. Tone Mapping) by using OpenCV. More to come soon!
- Bilateral Filter
- HDR Approximation Filter
- Toggleable Timer for Performance Analysis
A bilateral filter is a non-linear, edge-preserving, and noise-reducing smoothing filter for images. It replaces the intensity of each pixel with a weighted average of intensity values from nearby pixels. The overall effect of this filter is that sharp edges are preserved, whereas all other elements are smoothed-out.
This technique is used in image processing and computer graphics to map one set of colors to another to approximate the appearance of high-dynamic-range images in a medium that has a more limited dynamic range. The problem of strong contrast reduction from the scene radiance to the displayable range is addresses while preserving the image details and color appearance important to appreciate the original scene content.
The block size was selected at 256 which was found to give the best performance in GPU.
The block size was selected at 1024 which was found to give the best performance in GPU. You can see here that HDR tone mapping done on GPU using global memory takes more time to compute than the CPU. This shows how important the use of shared memory is to maximize the eficiency of GPU's parallelism.
When you are bound by your GPU's maximum number of threads per block: Messed-up color channel!! :P
