Hello,
I am working with C++(OpenCv) and I want to compute the runtime of my method to compare it with other methods. I use clock_t tStart = clock(); and printf("Time taken: %.4fs\n", (double)(clock() - tStart)/CLOCKS_PER_SEC);
The problem is that i don't know where to put the clock start? Is it after image reading and image preprocessing or after them? also the same for clock end.
Thank you
put the clock start after image reading. The time of reading image should not be counted. If the image preprocessing is a part of your algorithm, this time must be taken into accout. And place clock end at the end of your image processing algotithm. Not count the time of saving image or other time has nothing to do with your method.
You should definitely NOT count the loading and saving time into your computations time (or other methods time). Also, be sure to take the following into account:
You might also run a profiler like AQTime in order to find the slow spots of your algorithms and improve on them. Remember that algorithms in OpenCV are mostly implemented and optimized by professionals and that without profiling and improving your code, you might gest suboptimal results.
https://smartbear.com/product/aqtime-pro/overview/
http://stackoverflow.com/questions/67554/whats-the-best-free-c-profiler-for-windows
Thank you very much, dear Bruno Martin, for your suggestions. I use I profiler in code::blocks, it helps me to know which part can be parallelized.
If you have added your own preprocessing algorithms to give better results like Morphological operations, normalizing, thining, Skew correction then it is necessary to start calculating computation time before preprocessing algorithm. If there is no such algorithm then not necessary to do it before preprocessing do it after preprocessing.
If you are using profiler then AQtime Pro is a software performance exploration suite to help developers track down memory profiling issues, CPU and other I/O bottlenecks, perform comprehensive code coverage analysis, and perform fault simulation.
Actually, I used bilateral filter before applying my method to enhance results. Should I take in consideration the execution time of this filter?
If the results you get need this bilateral filtering in order to work (or work well) and the traditional methods don't, YES you must include this time also into your computation.
That's essential to the integrity of your publication... Otherwise you would be misleading the reader as to the real performance of all the operations needed to make your method work as you report them. It is even more important in the case of a relatively computationally intensive operation as a bilateral filter (we are not talking a simple averaging)!
We often see authors reporting the following benchmarks:
I suggest you have a look at the following links. The discuss how to make your publication more easily comparable and testable by other researchers, thereby increasing their thrust in your work and its potential spread.
https://hal.archives-ouvertes.fr/file/index/docid/577010/filename/ipol_iccs.pdf
https://hal.inria.fr/hal-01110206/file/reproducible-research-20150127.pdf
http://www.bitss.org/wp-content/uploads/2014/06/reproducible-research-for-scientific-computing-tools-and-strategies-for-changing-the-culture3.pdf
Thank you very much, dear Bruno Martin. Your answer is clear. My aim is to give real and trusty results which lead my conscience to be relieved.
in mat lab : put tic at processing start and toc at the end
tic
imshow(a)
toc
#ifndef TIMER_H
#define TIMER_H
#ifdef __linux__
#include
#else
#include
#endif
#include
double getTime(){
#ifdef __linux__
timeval tv;
gettimeofday(&tv, 0);
return tv.tv_sec + tv.tv_usec * 1e-6;
#else
return 1.0 * clock() / CLOCKS_PER_SEC;
#endif
}
#define tic tic_f(__LINE__);
#define toc toc_f(__LINE__);
#define MAX_LINE 10000
double _timer[MAX_LINE] = {0};
int _fromLine[MAX_LINE] = {0};
int _lastLineNum = -1;
double _lastRdtsc = 0;
inline void tic_f(int line){
_lastLineNum = line;
_lastRdtsc = getTime();
}
inline void toc_f(int line){
double t = getTime();
_fromLine[line] = _lastLineNum;
_timer[line] += t - _lastRdtsc;
_lastLineNum = line;
_lastRdtsc = t;
}
inline void tictoc(FILE *out){
int i;
for(i = 0; i < MAX_LINE; i++){
if(_timer[i] != 0)
fprintf(out, "line %d - %d: %.3f\n", _fromLine[i], i, _timer[i]);
}
fflush(out);
}
#endif
put the clock start after image reading. The time of reading image should not be counted. If the image preprocessing is a part of your algorithm, this time must be taken into accout. And place clock end at the end of your image processing algotithm. Not count the time of saving image or other time has nothing to do with your method.
See std::clock() function.
const clock_t begin_time = clock();
// do something
std::cout
you may calculate the time consumed by each step, then combine the total time for for the whole method, this would give a clearer vision for time consumed.
Dont calculate the time consumed by reading and saving the image so you start the timer at each step of your technique and end with it so you can know the exact time for each step.
The best way would be to use a tool like GPROF:
- http://www.thegeekstuff.com/2012/08/gprof-tutorial/
- http://people.sc.fsu.edu/~jburkardt/cpp_src/gprof/gprof.html
And you should remember to run exclusively a single program during profiling.
In Matlab use pair of functions tic for starting the timer of execution time and toc for stopping the timer and to display the time span between usage of tic and toc.
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing
- Image Processing