Hello,
I am using the Stitiching module and ROI to stitch two photos (968x1296 each).
On the iPhone 4 it takes more than 30 seconds (about 14 seconds to find features and 18 seconds to compose images + other calculations).
Can you please tell me if there is any way to speed up this process?
The first thing to check is whether you compile with full optimizations: set optimization level to -O2/-O3
If that doesn't help you enough, you should play with the stitcher parameters:
double registrationResol()
void setRegistrationResol(double resol_mpx)
double seamEstimationResol()
void setSeamEstimationResol(double resol_mpx)
double compositingResol()
void setCompositingResol(double resol_mpx)
They control the quality and the speed of your stitching process.
Third, you should try it on a Tegra3 device - that's an Android, btw ;)
Finally, if you still do not like the result, why not try to optimize it yourself more? Here is a starting point into image processing optimizations. But beware that you may cut your fingers. http://answers.opencv.org/question/755/object-detection-slow/#760
Hello Sammy. Thank you very much for your answer, it's very useful.
setRegistrationResol helped me to significantly(x2.5) speed up the process of feature detection. I changed the default value to 0.1 and if the search of features is unsuccessful, then I'm automatically changing it to the highest value of 1.0.
I have one more issue with "setCompositingResol" though.
I tried to change its value and when the value is "0.1" the stitching process is blazingly fast, but if I try to stitch the stitched photo that I get with this value with a regular photo, then "finding features" shows me an error no matter what's the value of "registrationResol". If I increase "compositingResol", then features can be found, but it requires much more time than it does with the default "compositingResol" value, so unfortunately I failed to decrease the stitching time here.
Maybe you have some ideas why this happens?
Asked: 2012-10-15 04:21:39 -0600
Seen: 2,485 times
Last updated: Oct 15 '12
Which features are you using? Did you try Orb or other features?
Yes, I tried Orb, it works fast, but sometimes it finds it difficult to detect enough features. I came to a conclusion that by changing registrationResol I get much better results with the default feature detector.