文件列表:

CMakeLists.txt (1672, 2013-08-08)
cudaImage.cu (3290, 2013-08-08)
cudaImage.h (870, 2013-08-08)
cudaSift.h (1133, 2013-08-08)
cudaSiftD.cu (27949, 2013-08-08)
cudaSiftD.h (3233, 2013-08-08)
cudaSiftD_old.cu (11158, 2013-08-08)
cudaSiftH.cu (15871, 2013-08-08)
cudaSiftH.h (3342, 2013-08-08)
cudaSiftH_old.cu (6668, 2013-08-08)
cudaSiftH_old.h (790, 2013-08-08)
cudaSift_gold.cpp (3098, 2013-08-08)
cudautils.h (2822, 2013-08-08)
data\left.pgm (1228816, 2013-08-08)
data\righ.pgm (1228816, 2013-08-08)
data\rimg_pts.pgm (1228816, 2013-08-08)
geomFuncs.cpp (2437, 2013-08-08)
mainSift.cpp (6190, 2013-08-08)
matching.cu (13243, 2013-08-08)
oldcode\cudaImage.cu (4146, 2013-08-08)
oldcode\cudaImage.h (920, 2013-08-08)
oldcode\cudaSift.h (1631, 2013-08-08)
oldcode\cudaSiftD.cu (33800, 2013-08-08)
oldcode\cudaSiftD.h (5585, 2013-08-08)
oldcode\cudaSiftH.cu (26557, 2013-08-08)
oldcode\cudaSiftH.h (3139, 2013-08-08)
oldcode\cudaSift_gold.cpp (3098, 2013-08-08)
oldcode\cudautils.h (2712, 2013-08-08)
oldcode\mainSift.cpp (11591, 2013-08-08)
oldcode\singular.cu (4390, 2013-08-08)
singular.cu (4388, 2013-08-08)
data (0, 2013-08-08)
oldcode (0, 2013-08-08)

This is the third version of a SIFT (Scale Invariant Feature Transform) implementation using CUDA for GPUs from NVidia. The first version is from 2007 and GPUs have evolved since then. This version is considerably faster than the previous versions and has been optimized for GeForce GTX 580. It should be reasonable fast also on more recent GPUs. On a GTX 580 GPU the code takes about 7 ms on a ***0x480 pixel image and 13 ms on a 1280x960 pixel image. The 2 ms and 1 ms respective of that is for data transfers inbetween GPU and CPU. If you keep data on the GPU things are thus somewhat faster. There is also code for brute-force matching of features and homography computation that takes about 7 ms and 4 ms for two sets of around 1500 SIFT features each. The code relies on CMake for compilation and OpenCV for image containers. OpenCV can however be quite easily changed to something else. The code can be relatively hard to read, given the way things have been parallelized for maximum speed. The dependencies between files in the repository look like this: mainSift.cpp -> cudaSift.h -> cudaImage.h cudaImage.h cudaImage.cu -> cudaImage.h cudaSiftH.cu -> cudaImage.h cudaSift.h -> cudaImage.h cudaSiftD.h cudaSiftH.h -> cudaImage.h cudaSiftD.cu The code is free to use for non-commercial applications. If you use the code for research, please refer to the following upcoming paper. M. Bjrkman, N. Bergstrm and D. Kragic, "Detecting, segmenting and tracking unknown objects using multi-label MRF inference", CVIU (accepted for publication)

近期下载者：

相关文件：

评论：[我要评论] [举报此文件]

收藏者：