.. _x264_vs_openh264_dataset:

libx264 vs libopenh264 Dataset
==============================

Compare the **libx264** and the **libopenh264** encoders.

Download
--------

.. code:: shell

    mendevi download x264_vs_openh264.db.xz.torrent

* `x264_vs_openh264.json.xz <https://gitlab.inria.fr/rrichard/mendevi/-/raw/main/dataset/x264_vs_openh264.json.xz?inline=false>`_ (18.3 Mo).
* `x264_vs_openh264.db.xz.torrent <https://gitlab.inria.fr/rrichard/mendevi/-/raw/main/dataset/x264_vs_openh264.db.xz.torrent?inline=false>`_ (7.0 Go).

Plots
-----

Distortion and Consumption as a function of bitrate
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Overall trend for all measures:

.. code:: shell

    mendevi plot x264_vs_openh264.db -x rate -y psnr -y vmaf -wx category -c '(encoder, profile)' -m effort

.. image:: /_static/media/x264_vs_openh264_bdr.svg
    :alt: Distortion as a function of bitrate

Result for a single video, including energy:

.. code:: shell

    mendevi plot x264_vs_openh264.db -x rate -x energy -y psnr -c encoder -m effort -wy profile -f "ref_stem=='park_joy'"

.. image:: /_static/media/x264_vs_openh264_cut.svg
    :alt: Link between distortion and energy for each encoder


Compression efficiency for PSNR=35dB
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

.. code:: shell

    mendevi plot x264_vs_openh264.db -x rate -y energy -c '(encoder, profile)' -m category -f '34.5<psnr<35.5'

.. image:: /_static/media/x264_vs_openh264_efficiency_psnr.svg
    :alt: Consumption as a function of bitrate for a psnr of 35dB

Compression efficiency for rate=2Mb/s
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

.. code:: shell

    mendevi plot x264_vs_openh264.db -x rate -y energy -c '(encoder, profile)' -m category -f '1.8e6<rate<2.2e6'

.. image:: /_static/media/x264_vs_openh264_efficiency_rate.svg
    :alt: Consumption as a function of bitrate for a bitrate of 2Mb/s

Compression efficiency for fixed energy
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

.. code:: shell

    mendevi plot x264_vs_openh264.db -x rate -y psnr -wx 'round(energy/nbr_frames, 1)' -c encoder -m '(profile, effort)' -f "0.58<energy/nbr_frames<0.62 or 3.95<energy/nbr_frames<4.05"

.. image:: /_static/media/x264_vs_openh264_efficiency_energy.svg
    :alt: Consumption as a function of bitrate for a fixed energy

Conclusion
----------

* The **main consumption factor is the resolution**, not the encoder.
* At equal resolution and quality, ``libopenh264`` consumes less energy than ``libx264``.
* ``libx264`` **has a large variance in energy consumption** depending on the preset and the video content. It is therefore much less predictable than ``libopenh264``.

Ressources
^^^^^^^^^^

`this paper <https://iopscience.iop.org/article/10.1088/1757-899X/1172/1/012036/pdf>`_ (Performance comparison among popular
implementations of H.264 encoders) compare the distortion of that codecs.

Reproduce
---------

.. code:: shell

    for video in ctc/*.mkv; do
        mendevi prepare $video -p sd
        mendevi prepare $video -p fhd
    done
    mendevi encode -e fast -e medium -e slow -m cbr -m vbr -c libx264 -c libopenh264 reference*
    mendevi probe --psnr --ssim --vif --vmaf sample*