World: r4wp
[#Red] Red language group
older newer | first last |
Kaj 27-Jun-2013 [8917] | im1: as-byte 0 |
Bo 27-Jun-2013 [8918x3] | Well, that seems to have worked around the problem! Thanks guys! |
Now I can work on making some real progress! :-) | |
I'm amazed by how fast Red/System is. | |
PeterWood 27-Jun-2013 [8921] | Excellent ! |
Kaj 27-Jun-2013 [8922] | Your code could be optimised a bit :-) |
james_nak 27-Jun-2013 [8923] | Kaj, I think there was something to the jarsigner being commented out in the file. I'll have to wait for Doc and Pekr to chime in. |
Kaj 27-Jun-2013 [8924] | Yes |
Bo 27-Jun-2013 [8925] | I'd love to learn how to optimize my code! |
Kaj 27-Jun-2013 [8926] | Are you low on memory? |
PeterWood 27-Jun-2013 [8927] | A simple one: im1: (as integer! r + g + b) / 3 |
Kaj 27-Jun-2013 [8928] | That won't work |
PeterWood 27-Jun-2013 [8929x2] | I meant : im1: (as integer! r) + (as integer! g) + (as integer! b) / 3 |
2 additions & 1 division against 3 divisions & 2 additions | |
Kaj 27-Jun-2013 [8931x2] | im1: (as-integer img1/r) + img1/g + img1/b / 3 |
But yes, that was my general idea | |
PeterWood 27-Jun-2013 [8933] | :-) |
Bo 27-Jun-2013 [8934] | I now have a working rudimentary motion detection algorithm. Now it's time to tune it with my special formula. |
Kaj 27-Jun-2013 [8935] | If you're not memory constrained, you can leave the / 3 out completely and compare two grayscales as integer values |
PeterWood 27-Jun-2013 [8936x3] | Im1 is already an integer in Bo's code. |
if you're desparate for a few millisecs, reversing the loop to start at the end and work toward the front will help. | |
(Subtraction being marginally faster then addiiton). | |
Kaj 27-Jun-2013 [8939] | That could be cast to byte again, to save the grayscale, but yes, there's no need to divide by 3 if it's just for the comparison |
Bo 27-Jun-2013 [8940] | I'm passing the processed image back out to Rebol to convert back into an image for visual verification of the motion detection algorithm. That part works fine. :-) |
Kaj 27-Jun-2013 [8941x3] | Subtraction faster than addition? That would surprise me. It's comparison to 0 that may be faster |
Further, the bumping of r, g and b each iteration is relatively much work, because you're also having to use them as indexes. It's more efficient to make a byte pointer for each and advance those | |
But it's probably even faster to iterate through the image with one integer pointer, because the pixels happen to be four bytes, and then use indexes of 2, 3 and 4 for r, g and b | |
Bo 27-Jun-2013 [8944] | So are you saying that if 'img1 is a pointer, to do the following? img1r: img1 + 2 img1g: img1 + 3 img1b: img1 + 4 and then img1r: img1r + 4 img1g: img1g + 4 img1b: img1b + 4 would be faster than what I'm doing now? |
PeterWood 27-Jun-2013 [8945] | You're correct. I checked and the optimisation benefit through "loop inversion" apparently comes from reducing the number of jumps in the underlying code not difference in speed in atrithmetic. |
Bo 27-Jun-2013 [8946] | OK, so this would be even faster: r: 2 g: 3 b: 4 until [ ...do stuff here... img1: img1 + 4 ...comparison... ] |
Kaj 27-Jun-2013 [8947] | Bo, yes, that would be more efficient, because you can then index with 1 |
Maxim 27-Jun-2013 [8948] | Bo, that's basically how it'd be done in C |
PeterWood 27-Jun-2013 [8949] | Kaj do you know if Red/System generates the same code for : img1/value and img!/1 ? |
Kaj 27-Jun-2013 [8950] | Yes, it's equivalent |
Bo 27-Jun-2013 [8951x2] | Thanks for the tips, guys! |
There is no way for Red/System to read a directory listing, is there? | |
Kaj 27-Jun-2013 [8953x2] | Perhaps fastest: pixel: as int-ptr! img1 + 1 ... rgb: as-binary pixel im1: (as-integer rgb/1) + rgb/2 + rgb/3 / 3 pixel: pixel + 1 |
That code would have to be tested, depends on lots of details | |
Bo 27-Jun-2013 [8955] | I'm only looping 20'000 times or so...not a lot for Red/System to handle. ;-) |
Kaj 27-Jun-2013 [8956] | But it's going to run on a Raspberry in realtime |
Bo 27-Jun-2013 [8957x2] | I was joking. :-) |
Can Red/System read a directory listing? | |
Kaj 27-Jun-2013 [8959x3] | It should, but you have to interpret the data manually |
I haven't looked into it yet | |
I see you'd need some extra definitions that aren't in my C library binding yet | |
Pekr 28-Jun-2013 [8962x2] | james & kaj: I have JDK installed too, you should be sure, that you can call java, javac from whatever dir = it is in the system lookup PATH. And - Rebol was downloading supporting tools, then it hang in console, but in fact, there was a dialog box hidden in the background, asking me for write permission. Unless you allow that, the tools are not going to be saved into requested directory ... |
My directory structure is: build\tools: \api\android.jar jli.dll aapt.exe keytool.exe zipalign.exe | |
DocKimbel 28-Jun-2013 [8964] | @james_nak, first delete any jarsigner exe you might have downloaded or copied outside of the java folders. Then try to run jarsigner from the Red folder, if it says it can't be found, then you have to add to your PATH environment variable, the path to the java/bin/ folder of your installed JDK. |
Pekr 28-Jun-2013 [8965] | Doc - how do your slides go? :-) I hope you have prepared many questions for Carl :-) |
DocKimbel 28-Jun-2013 [8966] | Barely started, gone to see "Man of Steel" yesterday night to get some inspiration. ;-) |
older newer | first last |