ffmpeg-encoding-benchmark/Sparks_in_Blender_encode.py

#!/usr/bin/env python3

import os

import time

import csv

import ffmpy

from multiprocessing import cpu_count

from collections import OrderedDict

import json

from typing import Union, Any

# encoding options used
encoding: dict[str, Any] = {
    "libx264": {
        "crf": [10, 14, 16, 18, 20, 22, 25],
        "presets": [
            "superfast",
            "veryfast",
            "faster",
            "fast",
            "medium",
            "slow",
            "slower",
            "veryslow",
        ],
    },
    "libx265": {
        "crf": [10, 14, 16, 18, 20, 22, 25],
        "presets": [
            "superfast",
            "veryfast",
            "faster",
            "fast",
            "medium",
            "slow",
            "slower",
            "veryslow",
        ],
    },
    "libaom-av1": {"crf": [20, 25, 30, 35, 40], "presets": [0, 2, 4, 6]},
    "libsvtav1": {"crf": [20, 25, 30, 35, 40], "presets": [0, 4, 8, 12]},
}

# program version
# make tests reproducible by tag
version = "v0.0.1a2"

# always round timestamp to integer
def now():
    return int(time.time())


def write_line(
    codec: str,
    crf: int,
    preset: Union[str, int],
    infile: str,
    outfilesize: float,
    enctime: int,
    vmafmean: float,
    vmafmin: float,
):
    """
    Write line to data csv

    Parameters:
        codec (str): Codec used
        crf (int): CRF used
        preset (str/int): Preset used
        infile (str): Input file name
        outfilesize (float): Size of output file
        enctime (int): Time to encode
        vmafmean (float): Mean VMAF score
        vmafmin (float): Min VMAF score
    """
    with open(datafile, "a", newline="") as file:
        write = csv.writer(file)
        write.writerow(
            (
                codec,
                crf,
                preset,
                infile,
                outfilesize,
                enctime,
                vmafmean,
                vmafmin,
            )
        )


def encode_general(
    inputfile: str, outputfile: str, codec: str, crf: int, preset: Union[str, int]
):
    """
    General encoding function

    Parameters:
    inputfile (str): Path to input file
    outputfile (str): Path to output file
    codec (str): Codec used
    crf (int): CRF value
    preset (str/int): Choosen preset
    """
    ff = ffmpy.FFmpeg(
        inputs={inputfile: None},
        outputs={
            outputfile: "-c:v {videocodec} -crf {crf} -preset {preset} -g 240 -map 0:v:0 ".format(
                videocodec=codec,
                crf=crf,
                preset=preset,
            )
        },
    )

    return ff


def encode_libaom(inputfile: str, outputfile: str, crf: int, preset: Union[str, int]):
    """
    Encoding with libaom

    Parameters:
    inputfile (str): Path to input file
    outputfile (str): Path to output file
    crf (int): CRF value
    preset (str/int): Choosen preset
    """
    ff = ffmpy.FFmpeg(
        inputs={inputfile: None},
        outputs={
            outputfile: "-c:v libaom-av1 -crf {crf} -b:v 0 -cpu-used {preset} -row-mt 1 -tiles 2x2 -g 240 -map 0:v:0 ".format(
                crf=crf,
                preset=preset,
            )
        },
    )

    return ff


def score_vmaf(outputfile: str, inputfile: str) -> dict[str, float]:
    """
    Calculate a file's VMAF score. Higher is better

    Parameters:
    outputfile (str): Path to output file
    inputfile (str): Path to input file

    Return:
    dict[str, float]: VMAF mean and min value
    """
    ff = ffmpy.FFmpeg(
        inputs=OrderedDict([(outputfile, None), (inputfile, None)]),
        outputs={
            "-": "-filter_complex libvmaf=log_fmt=json:n_threads={cputhreads}:log_path=vmaf.json -f null".format(
                cputhreads=cpu_count()
            )
        },
    )

    ff.run()

    with open("vmaf.json", "r") as file:
        vmafall = json.load(file)

    vmaf: dict[str, float] = {
        "mean": vmafall["pooled_metrics"]["vmaf"]["mean"],
        "min": vmafall["pooled_metrics"]["vmaf"]["min"],
    }

    return vmaf


def score_psnr(outputfile: str, inputfile: str):
    """
    Calculate a file's MSE (mean-square error) using PSNR. A lower value is better

    Parameters:
    outputfile (str): Path to output file
    inputfile (str): Path to input file

    Return:
    TBD
    """
    ff = ffmpy.FFmpeg(
        inputs=OrderedDict([(outputfile, None), (inputfile, None)]),
        outputs={"-": "-lavfi psnr=stats_file=psnr.log -f null"},
    )

    ff.run()

    # Steps to get mse value


def score_ssim(outputfile: str, inputfile: str):
    """
    Calculate a file's SSIM rating. TBD

    Parameters:
    outputfile (str): Path to output file
    inputfile (str): Path to input file

    Return:
    TBD
    """
    ff = ffmpy.FFmpeg(
        inputs=OrderedDict([(outputfile, None), (inputfile, None)]),
        outputs={"-": "-lavfi ssim=stats_file=ssim.log -f null"},
    )

    ff.run()

    # Steps to get ssim value


if __name__ == "__main__":
    if not os.path.isdir("encoded"):
        os.mkdir("encoded")

    datafile = version + "-data-" + str(now()) + ".csv"

    with open(datafile, "w", newline="") as file:
        write = csv.writer(file)
        write.writerow(
            (
                "Codec",
                "CRF",
                "Preset",
                "Input file",
                "Output file size (MiB)",
                "Encode time (s)",
                "VMAF Score (mean)",
                "VMAF Score (min)",
            )
        )

    inputfile = "Sparks_in_Blender.webm"

    for codec in encoding:
        for crf in encoding[codec]["crf"]:
            for preset in encoding[codec]["presets"]:
                outputfile = os.path.join(
                    "encoded",
                    (
                        "Sparks_in_Blender-codec_"
                        + codec
                        + "-crf_"
                        + str(crf)
                        + "-preset_"
                        + str(preset)
                        + ".mkv"
                    ),
                )

                # libaom needs additional options
                if codec == "libaom-av1":
                    ff = encode_libaom(
                        inputfile=inputfile,
                        outputfile=outputfile,
                        crf=crf,
                        preset=preset,
                    )
                else:
                    ff = encode_general(
                        inputfile=inputfile,
                        outputfile=outputfile,
                        codec=codec,
                        crf=crf,
                        preset=preset,
                    )

                # execute previously defined encoding settings
                starttime = now()
                ff.run()
                endtime = now()
                difftime = int(endtime - starttime)

                outputfilesize = os.path.getsize(outputfile) / 1024 / 1024

                vmaf = score_vmaf(outputfile=outputfile, inputfile=inputfile)

                write_line(
                    codec=codec,
                    crf=crf,
                    preset=preset,
                    infile="Sparks_in_Blender.webm",
                    outfilesize=outputfilesize,
                    enctime=difftime,
                    vmafmean=vmaf["mean"],
                    vmafmin=vmaf["min"],
                )

                os.remove(outputfile)