Interesting! As long as there's some easily measurable perf difference (like 2x
is fine), I should be able to figure out what's going on. Though it's weird
that your original image is more like 30x.
Let's try something else. Assuming your original image is not something you can
share, what about this:
oiiotool original.exr --mulc 0.0 -o black.exr
That should make the pixels black, but retain all the metadata, channel names,
and data types of the original that had a large performance difference. It
should be a small file (black compresses very well) and will obviously contain
no imagery that is secret.
In one "part" of an exr file, all channels are either deep or flat, so "deep"
is all or none of the channels in that part/subimage.
> On Nov 10, 2021, at 12:30 PM, Daniel Flehner Heen <flehnerhee...@gmail.com>
> wrote:
>
> Hi, Larry!
>
> Thanks for looking into this. I created an image with IBA.noise() in 70
> channels and there still is a performance difference.
> Not that big though: 2.3 seconds (python) vs 0.9 (oiiotool)
>
> Not sure if it makes a difference, but the slow image has a mix of float and
> half channels and is tiled 64x64.
> Don't know exactly if any of the channels are deep. Will ImageBuf.deep return
> True if any channel is deep?
> a buf.deep call returns False.
>
>
> On Wed, Nov 10, 2021 at 6:16 PM Larry Gritz <l...@larrygritz.com
> <mailto:l...@larrygritz.com>> wrote:
> Well that sure is weird! Usually, both the python bindings and the oiiotool
> operation are pretty thin wrappers around the corresponding ImageBufAlgo
> function which is doing almost all the work. A big perf difference is not
> expected.
>
> Can I assume that this is reproducible by any 70 channel image? Like I can
> just use oiiotool to generate a black image and see a big speed disparity
> like this? Or is it in any way specific to the particular image?
>
> Wednesdays are pretty packed, but I'l try to take a look at this today if I
> can.
>
>
>> On Nov 10, 2021, at 4:08 AM, Daniel Flehner Heen <flehnerhee...@gmail.com
>> <mailto:flehnerhee...@gmail.com>> wrote:
>>
>> Hi!
>>
>> I'm attempting to extract only the RGBA channels of an EXR with 70+
>> channels.
>>
>> Using:
>> oiiotool -i:ch=R,G,B,A /path/to/gigantic.exr -o manageable.exr
>> takes about 8 seconds.
>>
>> In python (2.7 !caugh..):
>> buf = oiio.ImageBuf('/path/to/gigantic.exr')
>> obuf = oiio.ImageBufAlgo.channels(buf, ('R', 'G', 'B', 'A'))
>> obuf.write('manageable.exr')
>> takes 4+ minutes
>>
>> I tried only extracting one channel and it took the exact amount of time. I
>> expect a performance hit using python but this seems a bit off. I suspect
>> the python bindings are reading all the channels even though I'm asking for
>> a specific few.
>> I might of course be going about this completely wrong, in which case a
>> pointer in the right direction would be highly appreciated :)
>>
>> Straight read and write of the same file is done quickly.
>>
>> Thanks!
>>
>> --
>> -Daniel
>> _______________________________________________
>> Oiio-dev mailing list
>> Oiio-dev@lists.openimageio.org <mailto:Oiio-dev@lists.openimageio.org>
>> http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org
>> <http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org>
>
> --
> Larry Gritz
> l...@larrygritz.com <mailto:l...@larrygritz.com>
>
>
>
>
> _______________________________________________
> Oiio-dev mailing list
> Oiio-dev@lists.openimageio.org <mailto:Oiio-dev@lists.openimageio.org>
> http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org
> <http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org>
>
>
> --
> -Daniel
> _______________________________________________
> Oiio-dev mailing list
> Oiio-dev@lists.openimageio.org
> http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org
--
Larry Gritz
l...@larrygritz.com
_______________________________________________
Oiio-dev mailing list
Oiio-dev@lists.openimageio.org
http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org
