OpenMP parallelization of the Hessian Affine detector with SIFT descriptor (original code here ). Important details about this project can be found in the report file in this repository.
This project has been tested on:
- Intel®Xeon Phi TM 7210, aka Knights Landing (KNL), 64 cores, with NUMA memory architecture.
- Intel(R) Core(TM) i7-5820K, 6 cores, DD3 Memory, 1536 kB L2 Cache, 15360 kB L3 Cache.
- C++ Intel Compiler 2017.3.191.
- OpenCV 3.2 commit 9053839
- OpenCV contrib commit f39f415
./hessaff image [maxThreads][nRuns][scability]
Where:
imageis the input image. The image extension must be supported by OpenCV.maxThreads(default: omp_get_max_threads()) is the upper bound of the number of threads used in the program execution. Look at scalability for more details.nRuns(defualt: 1) how many times each parallel configuration is executed. This is useful for evaluation purpose.scalability(default: 0) if 1 perform a scalability test, otherwise get the descriptors in image using maxThreads threads for nRuns times. In a scalability test, maxThreads different parallel configurations are executed: the first one uses omp_set_num_threads(1), the i-th configura- tion uses omp_set_num_threads(i) and the last one omp_set_num_threads(maxThreads). Each configuration is executed nRuns times.
A typical execution output is:
[spm1428@ninja hesaffCC]$ ./hesaff darkKnight.jpg
---------------------------------------------------
File: darkKnight.jpg
Using 64 threads
descriptor time 2.13762 seconds
parallel time=1.00586
descriptors.rows=41131
avg descriptor time=2.13762
avg parallel time=1.00586
- Movie Posters
- Paintings images
- Oxford Building Dataset
