SiftGPU is an implementation of SIFT [1] for GPU. Basically, GPU shaders are used in Gaussian pyramid construction, DoG keypoint detection and descriptor generation. Not only does SiftGPU process pixels and features paralelly with GPU, this implementation also builds compact feature list by using GPU reduction [3], which changes the per-pixel processing to per-feature processing and reduces readback time.
SiftGPU is inspired by Andrea Vedaldi's sift++[2] and Sudipta N Sinha et al's GPU-SIFT[4] . Many parameters of sift++ ( for example, number of octaves, number of DOG levels, edge threshold, etc) are also available in SiftGPU.
SiftGPU uses CG shaders to program GPUs. The shader programs are dynamically generated according to the parameters user specified. A working GLSL version is also available in the package.
A cg-based sift matcher, SiftMatchGPU, is included in this code package. It multiplies the descriptor matrix on GPU and find the closest feature matches. The guided matching using homoraphy/fundamental matrix is also supported.
The entire functionality works fully only on hardware that supports cg profile fp40/vp40 or higher, for example, nvidia 7900, 8800. If your GPU does not support fp40/vp40, orientation computation of SIFT will be simplified, edge elimination will be ignored, and descriptor will be ignored.
SiftGPU uses CG (1.5 or higher),
DevIL Image Library, GLEW and
GLUT.
You'll need to make sure that your system has all the dependening
libraries. SiftGPU should be able to run on any operation system that
supports the above libraries VC6 workspace and VS2005 solution are
provided as VC\SiftGPU.dsw and VC\SiftGPU.sln
.
Linux makefile is in folder Linux of the package and Xcode project is in folder Mac of the package
NOTE: To update the depending libaries, you'll need to update the header files in SiftGPU\Include\, and the windows libraries in SiftGPU\SiftGPU\Lib and SiftGPU\TestWin\Lib.
The whole package SiftGPU-V320.zip
(5.9MB; Including code, manual, windows binary
and some test images)
|
Example of using siftgpu is in file SiftGPU\TestWin\src\SimpleSIFT.cpp. A complete
list of
changes in all previous version can be found
here, and old versions of
siftgpu can be found through this link. |
|
What's New in V319-V320 |
You might also be interest in SiftGPU in Matlab ported by Adam Chapman
Below is the comparision with Lowe's SIFT on box.pgm using the comparision code from Vedaldi's SIFT .
Below are some of our speed experiment results on V311. More complete experiment results can be found here. (The experiment images of different sizes are all resized from this image)
The experiment below is the comparision of the new version and the old version.
[1] D. G. Lowe. Distinctive image features from
scale-invariant keypoints . International Journal of Computer Vision, November 2004.
[2] A. Vedaldi. sift++, http://vision.ucla.edu/~vedaldi/code/siftpp/siftpp.html .
[3] G. Ziegler, et al. GPU point list generation through histogram pyramids. In Technical Report, June 2006.
[4] Sudipta N Sinha,
Jan-Michael Frahm, Marc Pollefeys and Yakup Genc, "GPU-Based Video Feature
Tracking and Matching ",
EDGE 2006, workshop
on Edge Computing Using New Commodity Architectures, Chapel Hill, May
2006