Skip to main content

Read from and seek into video files as if they were Python sequences of PIL.Image-s.

Project description

https://travis-ci.org/rjw57/videosequence.svg?branch=master https://coveralls.io/repos/github/rjw57/videosequence/badge.svg?branch=master https://img.shields.io/pypi/v/videosequence.svg

Quite often I find myself writing scripts which need to load a few frames from a video file, process them and save the result to disk. It’s a pain to implement video opening, seeking and decoding over and over again and complex Python bindings are a little overkill for my needs.

Videosequence is a library which hides the complexity of simply opening a video file in Python as a sequence of images. It exposes a video file as just that: a Python sequence type containing PIL Image-s.

For example, suppose you want to dump every frame from a video stored in foo.mp4 starting from frame 100:

from contextlib import closing
from videosequence import VideoSequence

with closing(VideoSequence("foo.mp4")) as frames:
    for idx, frame in enumerate(frames[100:]):
        frame.save("frame{:04d}.jpg".format(idx))

You can load a single frame from a sequence just as easily. Let’s dump the final frame to another JPEG:

from contextlib import closing
from videosequence import VideoSequence

with closing(VideoSequence("foo.mp4")) as frames:
    frames[-1].save("final-frame.jpg")

In general, the VideoSequence behaves as if it were a long list of each frame in the video.

What VideoSequence does

  • Frame-accurate seeking

  • Single frame indexing (vs[0], vs[-4], etc.)

  • Querying the length of the video (len(vs))

  • Slicing a sequence of frames (vs[100:], vs[-20:], vs[10:20], vs[::2], etc.)

  • Frames are represented as RGB PIL Image objects.

  • Can interoperate with numpy. E.g. np.asarray(vs[0]).

What VideoSequence does not

  • Handle files without exactly one (and only one) video stream

  • Audio

Caveats

  • Iterating forward one frame at a time is fast. Tricks such as iterating backwards or skipping n frames at a time work but is likely to be slow.

  • The implementation is based on GStreamer and so de facto only works on a modern Unix-alike such as Linux or FreeBSD.

  • The PyGObject introspection libraries must be installed. (See below.)

Installing

See the sections below for any OS-specific instructions. VideoSequence can be installed from the PyPI:

$ pip install --user videosequence

It can also be installed directly from git:

$ pip install --user git+git://github.com/rjw57/videosequence

Ubuntu and Debian

To install the Python GObject bindings:

$ sudo apt install gir1.2-gstreamer-1.0 gir1.2-gst-plugins-base-1.0 \
                   python-gi python3-gi

GStreamer is almost certainly already installed if you’ve got some modern desktop environment. If not:

$ sudo apt install libgstreamer1.0-dev gstreamer1.0-plugins-good

Contributing

Bug fixes and ports to new backends welcome. Please make sure that the tests still pass via tox before opening a new pull request. New functionality should come with tests, please.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

videosequence-1.1.0.tar.gz (5.6 kB view details)

Uploaded Source

File details

Details for the file videosequence-1.1.0.tar.gz.

File metadata

File hashes

Hashes for videosequence-1.1.0.tar.gz
Algorithm Hash digest
SHA256 062e4581f99ea70de42e1907ac87da49f419b91c6ce68e320901109578d3dd7b
MD5 0e61021da4b42dbf8c9fd62b36f32266
BLAKE2b-256 0de51888f72156e5d0625283c3d1586a72a58b7ddd715f29247ea10e63af7201

See more details on using hashes here.

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page