I use this action extensively throughout my routines and what I've noticed is that roughly 1 to 2 weeks ago, image recognition slowed down by orders of magnitude from approximately an average of 300 ms to 1-7 seconds.
In order to see if the culprit is perhaps the new zoom-friendly functionality that you added as part of feature request "Enhance "Find/Search Image on Screen & Move Mouse" action to recognize the image when it's zoomed in/out (e.g. by browser zoom level)", I downloaded the build from September 28 (6.868) and sure enough when I executed a routine with six image recognition actions, it executed in the blink of an eye. I repeated the test several times, because there are the vestiges of a half decent scientist within me somewhere, and obtained the same result. I then put the new version back and sure enough the image recognition slowed down dramatically as mentioned earlier.
I won't pretend to know what to suggest to you in terms of remedial action, I just wanted you to know that the existing action has been severely degraded. Going back precisely on what I just said, I for one one would appreciate a new build with the new Zoom-friendly functionality hidden behind a config flag or similar.
that is weird, they it still seems to be near instant here. You didn't change the matching method to one of the newly added ones, correct? (because they might be slower).
I'll check whether anything could cause it to use one of the new algorithms by accident
I just ran my test suite for this across both the old build and new build. On my machine there is no difference in speed if the same algorithm is used.
I also checked the code, both versions execute exactly the same code if the "Template Matching" algorithm is selected, so I guess the issue can't really be related to the "Find/Search Image" action.
Hi andreas, thanks once more for going above and beyond! I'll be honest with you, I didn't even notice the template buttons presence in the newest version! Which in turn means that I definitely did not twiddle that setting. Let me repeat your experiment with a build that's about a month old and I'll get back to you. That said, I think the probability of there being a problem, given the rigorous testing that you performed, is very low now - it's probably some issue with my machine or something.
Hi Andreas, I have some brief feedback from my usage of the template matching stuff that you implemented so quickly. In a nutshell, for the type of image that I need recognising (which is essentially 50 pixel by 50 pixel sized video game icons like the attached, only SIFT works.
Choosing any of the others fails completely to recognise the image, even if I literally just took a screen grab of it. Sample images attached, cheers and thanks for this enhancement, my macros are much nicer for it.
