kanryu / pipeffmpeg

A frontend for ffmpeg using only pipes, not under GPL, but under BSD license.

PipeFFmpeg

A frontend for ffmpeg(libav) using only pipes, to encode/decode/convert videos by each frame, not under GPL, but under BSD license.

Features:

• To get ffmpeg's infomation (versions,codecs,formats,pix_fmts,...)
• To decode frames from a video
• To encode a video within posting frames in Python

You can install and test pipeffmpeg as follows:

1. download ffmpeg(libav) binary or source and build that (not necessary as shared-library).
2. set bin/ffmpeg or ffmpeg.exe(under Windows) into PATH environment
3. download pipeffmpeg.py into your current directory or site-packages

pipeffmpeg (currently version 0.1.1) is also available at PyPI

pip install pipeffmpeg

Please report issues as well as configuration details for your working PipeFFmpeg installations at [email protected]

Thus, your feedback is welcome at [email protected].

pipeffmpeg's Issues

Reading video file halts after a while

After installed, I found it can't manage larger files, it will generally suspend reading after reading 1k frames.

import pipeffmpeg

vid = pipeffmpeg.InputVideoStream()
vid.open("A Long Video.mp4")

for i, img in enumerate(vid.readframe()):
    print i

subprocess PIPE objects have finite buffer size

Hello,

Me again. I've been fiddling with pipeffmpeg to get it working, and I've run across a reasonably large problem. subprocess PIPE objects have a limited buffer size, and once this buffer is overrun, the process will simply hang (without error). Basically, this limits the allowable size of videos that can be accessed via this strategy. Details on the larger issue can be found here.

A different strategy is to read the frames one at a time, or in cached batches. I've taken this strategy in https://github.com/davidcox/pipeffmpeg/tree/indexable_not_iterable

Basically, it computes time offsets from the video meta data and uses ffmpeg's -ss and -vframes options to pull out small batches of frames.

Thanks,
Dave