Log file Viewer

[17:06:51] <raincomplex> greetings

[17:07:19] <raincomplex> i'm having trouble finding information on image modes, in particular the I mode

[17:07:49] <wiredfool> I is not well documented

[17:07:55] <raincomplex> in the docs it says it's 32-bit signed integer pixel values, but it appears i'm only getting values 0-255

[17:07:58] <wiredfool> what do you need to know

[17:08:21] <wiredfool> how are you making/gettin an I image?

[17:08:38] <wiredfool> convert doesn't expand the range

[17:09:13] <raincomplex> img = PIL.Image.open('image.jpg'); i = img.convert('I'); pa = i.load(); pa[x,y]...

[17:09:25] <raincomplex> what do you mean about convert, that sounds like it's probably my issue

[17:09:39] <wiredfool> so what convert does is essentially an int cast

[17:10:06] <wiredfool> if you're going bigger, it's just storing an char in an int

[17:10:13] <wiredfool> if you're going smaller, it truncates

[17:10:20] <raincomplex> so i'm only getting one channel?

[17:10:33] <wiredfool> (after going grayscale)

[17:10:55] <wiredfool> I is single channel gray

[17:11:17] <raincomplex> oh

[17:11:47] <raincomplex> is there a way to access the RGB as a 32-bit integer instead of tuples like in 'RGB' mode?

[17:12:18] <raincomplex> i'll preface this by saying that i'm just looking for a way to make this a little faster

[17:12:24] <wiredfool> ah

[17:12:31] <raincomplex> and there's probably an easier way to do that than this

[17:12:35] <wiredfool> probably

[17:12:46] <wiredfool> step back a bit, what are you trying to do

[17:14:17] <raincomplex> twiddle bits in an image

[17:14:41] <raincomplex> steganography type deal

[17:15:00] <raincomplex> post on reddit got me thinking about it, and i'd not used PIL before

[17:15:09] <raincomplex> (usually i use pygame for this sort of thing)

[17:16:15] <wiredfool> well, you can get the bytestream with tobytes()

[17:16:47] <wiredfool> there's also imagemath

[17:17:25] <wiredfool> generally, the fast path is letting the c layer do the looping over the pixels in the image

[17:18:05] <raincomplex> unfortunately the access path is random and determined by python code

[17:18:11] <wiredfool> alternately, if you use pypy and cffi and the HEAD version, the pixel access object is way faster

[17:18:28] <wiredfool> tho, I don't remember if it's significantly faster than in the c version

[17:24:47] <wiredfool> the other thing that might help is to dump it into numpy, e.g. im.toarray(), and then use numpy to work on it.

[17:39:02] <raincomplex> hm. profiling it, it seems like there's a fair amount of overhead just from the number of function calls

[17:39:17] <raincomplex> maybe i should let speed go for this, haha

[17:39:26] <raincomplex> can always have a fast version in C

[18:22:10] <raincomplex> thanks for the help wired :D