A CUDA implementation of DAISY descriptor
- A CUDA implementation of DAISY descriptor based on [1,2,3].
- [1] E. Tola, V. Lepetit, and P. Fua. Daisy: an Efficient Dense Descriptor Applied to Wide Baseline Stereo. IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), 32(5):815–830, 2010.
- [2] https://github.com/etola/libdaisy
- [3] https://github.com/opencv/opencv_contrib/blob/master/modules/xfeatures2d/src/daisy.cpp
- OpenCV with cv::cuda
- CUDA
- OpenCV xfeatures2d module (optional)
- Maximum descriptor size is 256
- Supports only full scope (dense) mode
$ git clone https://github.com/gishi523/CU-DAISY.git
$ cd CU-DAISY
$ mkdir build
$ cd build
$ cmake ../
$ make
- If you want to compare with OpenCV DAISY, use
cmake -DWITH_OPENCV_DAISY=ON
./cudaisy image
./cudaisy ./images/leuven/img1.png
- Setup
- CPU : OpenCV DAISY with Core-i7 6700K(4.00 GHz/4Core/8T)
- GPU : CU-DAISY with GeForce GTX 1080
- Image size : 900 x 600 pixel
- Descriptor size : 200
| Parameters | CPU[msec] | GPU[msec] | Speed Up |
|---|---|---|---|
| Normalize:FULL, Interpolation:OFF | 178.5 | 8.1 | 22.2 |
| Normalize:FULL, Interpolation:ON | 234.2 | 9.0 | 26.0 |
| Normalize:PARTIAL, Interpolation:OFF | 172.7 | 8.1 | 21.5 |
| Normalize:PARTIAL, Interpolation:ON | 235.0 | 10.0 | 23.5 |
gishi523