Version: 3.2.x

Introduction

This project is also named as "FFmpeg-Encoder-Decoder-for-Python". It is implemented based on FFMpeg, Python-C-API and C++11. It is under GPL v3 License, and recommended for researching purposes.

With this package, users could:

Make use of all FFMpeg video codec

When decoding a video (or an online stream), like the original FFMpeg (C version), the provided APIs could detect the video format and codec format automatically. When encoding a video, users could control the codec format, bit rate and some other options by setting parameters.

Work with FFMpeg directly

This project invokes the FFMpeg C APIs in the bottom level. Unlike ffmpeg-python and pyffmpeg, our project is not driven by the FFMpeg CLI interfaces. The data format used by this package is np.ndarray. In other words, our project enables users to combine Numpy and FFMpeg directly.

Frame-level APIs

Unlike pyffmpeg, this package is not a simple wrapper of FFMpeg. Users could works on the frame-level APIs. For example, when decoding a video, users could get the data frame-by-frame. Each frame is a 3D np.ndarray.

Pre-compiled package

This package has been pre-compiled by the author. If users download the dependent dynamic libraries (.so or .dll), they do not need to compile the package by themself.

However, users could not work with this project in such cases:

Platform limited

Currently, we only support Linux and Windows. The Linux release is pre-compiled on Debian. It has been only tested in Ubuntu, Debian and Windows. In other cases, the pre-compiled library may not work. Users may need to compile the package by themselves.

Version limited

Currently, our project works with FFMpeg 4.4 and 5.0. Users need to download the dependent dynamic libraries to make the package work. The pip version is able to download the libraries automatically. The legacy versions of this project supports FFMpeg 3.3, 3.4.2 and 4.0. However, the legacy built packages are not technically supported now.

Audio not supported

Although the original FFMpeg supports both video and audio streams, our project only works on video streams. For example, if a video contains audio streams, our package would omit all audio frames in the bottom level. In other words, you could not perform audio analysis now. In the future (v4), we may support the audio frame analysis.

Filters not supported

Although the original FFMpeg supports some video processing tools (avfilter and postproc), our implementation drops these modules. Instead, we suggest that users should process the frames with pillow or openCV. On the other hand, our implementation still supports frame scaling and re-sampling (supported by swscale and swresample).