Coverage remained the same when pulling bef08fe on danielballan:video-guide into 5b07567 on scikit-image:master.

I made some changes in response to comments by @tacaswell and @sciunto and rebased.

Coverage remained the same when pulling 2ab42eb on danielballan:video-guide into 3814df2 on scikit-image:master.

@stefanv, this is probably ready to merge, pending any additional remarks from @sciunto or @tacaswell. More could be added, and will surely need to be updated in time, but as it stands it is a good, brief review of the available tools.

Changes Unknown when pulling 5d1d28f on danielballan:video-guide into * on scikit-image:master*.

What I'm still missing are command lines for doing the conversion, a la http://matplotlib.org/faq/howto_faq.html#make-a-movie

@danielballan What do you think about that suggestion?

Absolutely; it's good suggestion. This should enable users to get going quickly and easily. I will find time for it shortly.

Thanks, I very much appreciate you writing this up.

Coincidentally, I just deleted that section of the FAQ (not merged yet) for being terribly out of date (as it does not mention the MovieWriter classes).

Note by @cgohlke on the mailing list:

In case you are using Windows, try vidsrc 
<http://www.lfd.uci.edu/~gohlke/code/vidsrc.cpp.html>, which reads frame 
data of video files via
Microsoft's DirectShow IMediaDet interface. Binaries are included in the 
vLFD package installer at <http://www.lfd.uci.edu/~gohlke/pythonlibs/#vlfd>

from vidsrc import VideoSource
video = VideoSource("test.avi")
for frame in video:
     print(frame.mean())

Since that section of the mpl docs cannot be used any more, should we show how to split a movie into individual frames from the command line using ffmpeg and vlc?

Since the SciPy conference, PyAV has got better packaging and py3 compatibility, through my efforts and others'. On Linux/OSX, I think that is what we should recommend.

The latest release of PIMS wraps PyAV in this file. For scientific applications, we generally require random access by frame number, as opposed to random access by (approximate!) time. The PIMS wrapper builds a "table of contents" by scanning each packet. I am pretty sure it is the fastest possible accurate solution. Something like this should work its way into scikit-image in time. For now I can write up a minimal example that uses PyAV directly and a minimal example using PIMS.

As for Windows, I was not aware of Golke's video code and I'm very glad to know about it. For consistent results across platforms, I'd still like to see PyAV working on Windows someday, and I think it's possible. A good first step would be to get conda FFmpeg builds (with shared linking). Currently, even the Linux conda recipe is buggy.

But for the present: I will rewrite recommending PyAV or PyAV + PIMS for Linux/OSX and Golke's code for Windows. Perhaps we can package it on PyPI. Sound good?

That sounds like a plan to me! Will you guys take this one further--I'm
afraid my knowledge here is too little to contribute anything meaningful.

@danielballan, would you like me to package vidsrc the same way I did tifffile?

BTW, I've used VideoCapture in the past with success.

Yes, please do! I will crank out conda builds as well.

On Tue, Oct 14, 2014 at 3:19 PM, Steven Silvester [email protected]
wrote:

BTW, I've used VideoCapture http://videocapture.sourceforge.net/ in the
past with success.

—
Reply to this email directly or view it on GitHub
#1012 (comment)
.

Hmm, not as easy as I'd hoped. Compilation relies on a hard-coded path to the STRMBASE library from DirectShow.

I tried installing DirectX 9.0 SDK and the Windows SDK and could not find the libraries referenced in the source file. My goal was to distribute as a binary wheel, but I'm about fed up in general with Windows...

@holtzhau I think you two should start a support group.

Yeah I'm about ready to toss my Windows install disks and fold my arms.

@holtzhau Had the dubious honour of recently freezing a scikit-image and OpenCV application for distribution on Windows.

Hahaha, please! Sometimes I stare balefully at binstar search results hoping for rescuer.

Aha! I was not aware of the wheel convert command until just now. I will host the source code and a wheel file for vidsrc on PyPI. wheel convert - "Convert egg or wininst to wheel"

Yes, the wheel includes ffmpeg. Here's the repo that does the build:

https://github.com/MacPython/PyAV-wheels

It works the same way as the matplotlib wheel builds - the repo builds the library dependencies that OSX does not have by default, then uses the delocate utility to copy the required libraries into the package directory tree, and relink them.

I don't use conda myself because it seems unwise to make conda / Anaconda the only standard way of installing on OSX, but luckily for me, pip and virtualenvs are a pleasure to use on OSX now, because all the packages I install regularly have wheels.

I opened an issue for the automated wheel building on PyAV : PyAV-Org/PyAV#89

Thanks for explaining. This is great! The more working solutions, the better!

Matthew, this is excellent. Is there any progress on a delocate for Linux?

Daniel, are we about ready to merge here?

No, no progress on a delocate for Linux I'm afraid.

I get so busy...

If someone would like to help, I will try and carve out more time.

@stefanv I just pushed an update with more details and new links to build recipes and benchmarks. I also included brief installation instructions for each package, since packaging difficulties are a major consideration here.

With one more review, this should be ready to go.

Thanks @danielballan, this looks great. Some failures in building the docs, but as soon as those are fixed we can merge.

Travis is happy. Appveyor looks like it's failing on an unrelated issue with scipy.

Thank you very much for this excellent guide @danielballan; merged in 448230d

@stefanv see https://travis-ci.org/scikit-image/scikit-image/jobs/211736423:

Warning, treated as error:
/home/travis/build/scikit-image/scikit-image/doc/source/user_guide.rst:4: WARNING: toctree contains reference to nonexisting document u'user_guide/video'
make[1]: *** [html] Error 1
make[1]: Leaving directory `/home/travis/build/scikit-image/scikit-image/doc'
make: *** [html] Error 2

^ Should be fixed with c1d5320.

Thanks for the fixup, @soupault.

@danielballan not at all.
By the way, I've recently come across http://www.scikit-video.org/stable/ . Maybe we should add it also.

+1. It's extremely valuable to have a page summarizing the shooting range of the different libraries.