Hi Steven,
Here's the SO question:
https://stackoverflow.com/questions/73307400/extend-inner-array-in-3d-numpy-array
My initial code was this:
copy_channel_indices = [1, 3]
out_data = np.empty((height, width, 5), dest_data.dtype)
for y in range(height):
for x in range(width):
out_data[y][x][:3] = dest_data[y][x]
out_data[y][x][3] = src_data[y][x][1]
out_data[y][x][4] = src_data[y][x][3]
I got a 100x speed up by changing this code to this:
out_data = np.concatenate((dest_data, src_data[:, :, copy_channel_indices]),
axis=2)
Thanks
Hugh
On 18/08/2022 09:05, Steven Parker wrote:
Hi Hugh,
Perhaps you can share how you achieved the speedup from a tip on SO?
I'm doing something similar to you for a project, and hitting similar
performance issues.
I was initially doing the concatenating by hand in Python, which
was horrendously slow (10s for a UHD image), but was pointed in
the right direction by SO on that, which reduced it to <1% of the
time.
On Mon, Aug 15, 2022 at 12:54 PM Hugh Macdonald
<hugh.macdon...@scanlinevfx.com
<mailto:hugh.macdon...@scanlinevfx.com>> wrote:
Thanks Larry,
It's good to know that I wasn't missing anything OIIO-specific in
there. I was initially doing the concatenating by hand in Python,
which was horrendously slow (10s for a UHD image), but was pointed
in the right direction by SO on that, which reduced it to <1% of
the time.
That's helpful to see how I'd do it with ImageBuf. I think I'll
probably stick with what I've got for now, though, if there's
nothing majorly wrong with it, as it certainly works.
Thanks
Hugh
On 15/08/2022 19:36, Larry Gritz wrote:
Yeah, that looks like it should work.
If you really needed it done with ImageBuf specifically for some
reason, I think it might look like
# read old images
src_img = ImageBuf(src_image_file)
dst_img = ImageBuf(dst_image_file)
# make spec to describe the new image and allocate it
spec = src_img.spec.copy()
spec.nchannels = spec.nchannels + 1
spec.channel_names.append("new_layer.new_channel")
new_img = ImageBuf(spec)
# copy the right parts to the right places
# ... the first three channels get pasted from dst
ImageBufAlgo.paste(new_img, 0, 0, 0, 0, dst_img)
# ... the last channel gets pasted from src.G
roi = src.roi.copy()
roi.chbegin = 1
roi.chend = 2
ImageBufAlgo.paste(new_img, 0, 0, 0, spec.nchannels - 1, src_img,
roi)
# write results
new_img.write(dest_image_file)
(I haven't tried this, it's off the top of my head, but it's
probably close to correct.)
I don't necessarily think this is any better or more efficient
than what you did, I just thought it might be helpful to compare
how it would be done with ImageBuf alone, no ImageInput/ImageOutput.
-- lg
On Aug 15, 2022, at 9:14 AM, Hugh Macdonald
<hugh.macdon...@scanlinevfx.com
<mailto:hugh.macdon...@scanlinevfx.com>> wrote:
Hi Larry,
(b) is the closest to what I'm trying to do.
Since sending my initial email, I'm now doing this. This keeps
RGB in dest_img, and copies G from src_img into
new_layer.new_channel
src_img = oiio.ImageInput.open(src_image_file)
dest_img = oiio.ImageInput.open(dest_image_file)
src_data = src_img.read_image()
dest_data = dest_img.read_image()
src_spec = src_img.spec()
dest_spec = dest_img.spec()
src_img.close()
dest_img.close()
width = dest_spec.width
height = dest_spec.height
new_channels = list(dest_spec.channelnames)
new_channels.append("new_layer.new_channel")
copy_channel_indices = [1]
out_data = np.concatenate((dest_data, src_data[:, :,
copy_channel_indices]), axis=2)
out_img = oiio.ImageOutput.create(dest_image_file)
out_spec = oiio.ImageSpec(width, height, len(new_channels), "float32")
out_spec.channelnames = tuple(new_channels)
out_img.open(dest_image_file, out_spec)
out_img.write_image(out_data)
out_img.close()
Thanks
Hugh
On 14/08/2022 23:02, Larry Gritz wrote:
Hi, Hugh. Can you clarify whether you want to
(a) make a new RGB image R from B but G and B from A?
(b) make a new image with an additional channel, yielding {R,
G, B, and New}, where R,G,B came from A and New is the R
channel from B?
?
On Aug 8, 2022, at 8:14 AM, Hugh Macdonald
<hugh.macdon...@scanlinevfx.com
<mailto:hugh.macdon...@scanlinevfx.com>> wrote:
Hey all,
I'm not overly familiar with the OIIO python API, so apologies
if this is quite a simple question! I can't find any
information online on how to do this with the Python API...
I've got 2 EXRs, both with RGB channels, and I'd like to take
the R channel from one of them and put it into another channel
in the other.
I've been looking at doing a 2-step process. The first step
being to rename the channels in one ImageBuf, and to remove
the unneeded channels, and the second being to use
ImageBufAlgo.channel_append() to combine the two images together.
This feels like it is unnecessarily 2 steps. Could someone
point me at some examples for how to do this in a single step?
I'm sure that this is pretty simple, but all the examples that
I can find are for how to do this kind of thing using
oiiotool, rather than doing it directly using the API.
Thanks
--
Hugh
_______________________________________________
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
--
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
--
Hugh Macdonald
Product Owner, Volumetric Capture
Eyeline Studios
Ext: 4165
www.scanlinevfx.com <http://www.scanlinevfx.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
--
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
--
Hugh Macdonald
Product Owner, Volumetric Capture
Eyeline Studios
Ext: 4165
www.scanlinevfx.com <http://www.scanlinevfx.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
_______________________________________________
Oiio-dev mailing list
Oiio-dev@lists.openimageio.org
http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org
--
Hugh Macdonald
Product Owner, Volumetric Capture
Eyeline Studios
Ext: 4165
www.scanlinevfx.com
_______________________________________________
Oiio-dev mailing list
Oiio-dev@lists.openimageio.org
http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org
