def setUp(self) -> None:
super().setUp()
vm = video_meta_data(duration=3600.0, width=640, height=360)
am = audio_meta_data(duration=3600.0)
self.source = inputs.input_file(
'source.mp4',
inputs.Stream(VIDEO, vm),
inputs.Stream(AUDIO, am))
self.logo = inputs.input_file(
'logo.png',
inputs.Stream(VIDEO, video_meta_data(width=64, height=64)))
vm = video_meta_data(duration=10.0, width=640, height=360)
am = audio_meta_data(duration=10.0)
self.preroll = inputs.input_file(
'preroll.mp4',
inputs.Stream(VIDEO, vm),
inputs.Stream(AUDIO, am))
self.video_codec = X264(bitrate=3600000)
self.audio_codec = AAC(bitrate=192000)
self.output = outputs.output_file(
'output.mp4',
self.video_codec,
self.audio_codec)
self.ffmpeg = FFMPEG()
def test_detect_concat_buffering(self):
"""
When single source is used for multiple outputs, and one of outputs
has a preroll, buffering occurs, because to output first frame for a
non-preroll output, we need to buffer all preroll frames.
"""
cases = [
(False, True, True), # preroll + source / preroll + source
(False, True, False), # preroll + source / preroll
(True, False, True), # preroll + source / source
]
for case in cases:
with self.subTest(case):
raises, split_pre, split_src = case
ff = FFMPEG()
v1 = inputs.Stream(VIDEO, self.preroll.streams[0].meta)
a1 = inputs.Stream(AUDIO, self.preroll.streams[1].meta)
v2 = inputs.Stream(VIDEO, self.source.streams[0].meta)
a2 = inputs.Stream(AUDIO, self.source.streams[1].meta)
ff < inputs.input_file('preroll.mp4', v1, a1)
ff < inputs.input_file('source.mp4', v2, a2)
vf1 = v1 | filters.Split(VIDEO, output_count=int(split_pre) + 1)
vf2 = v2 | filters.Split(VIDEO, output_count=int(split_src) + 1)
af1 = a1 | filters.Split(AUDIO, output_count=int(split_pre) + 1)
af2 = a2 | filters.Split(AUDIO, output_count=int(split_src) + 1)
vc1 = vf1 | filters.Concat(VIDEO, input_count=2)
vf2 | vc1
ac1 = af1 | filters.Concat(AUDIO, input_count=2)
af2 | ac1
vc2 = filters.Concat(VIDEO, int(split_pre) + int(split_src))
if split_pre:
vf1 | vc2
if split_src:
vf2 | vc2
ac2 = filters.Concat(AUDIO, int(split_pre) + int(split_src))
if split_pre:
af1 | ac2
if split_src:
af2 | ac2
o1 = outputs.output_file("o1.mp4", X264(), AAC())
o2 = outputs.output_file("o2.mp4", X264(), AAC())
vc1 > o1
ac1 > o1
vc2 > o2
ac2 > o2
ff > o1
ff > o2
try:
ff.check_buffering()
except BufferError as e:
self.assertTrue(raises, e)
else:
self.assertFalse(raises)
def test_shortcut_outputs_with_codec(self):
""" Check ff > output shortcut if codecs list specified."""
ff = FFMPEG(input=inputs.input_file("input.mp4"))
scaled = ff.video | filters.Scale(width=1280, height=720)
with self.assertRaises(RuntimeError):
codec = codecs.VideoCodec("libx264")
out = ff > outputs.output_file("output.mp4", codec)
# at this moment codec is connected to ffmpeg input stream directly
# so scaled video stream could not be connected to output
scaled > out
codec = codecs.VideoCodec("libx264")
out = scaled > outputs.output_file("output.mp4", codec)
ff > out
def test_fix_trim_buffering(self):
"""
Trim buffering could be fixed with multiple source file deconding.
"""
ff = FFMPEG()
v1 = inputs.Stream(VIDEO, self.source.streams[0].meta)
a1 = inputs.Stream(AUDIO, self.source.streams[1].meta)
v2 = inputs.Stream(VIDEO, self.source.streams[0].meta)
a2 = inputs.Stream(AUDIO, self.source.streams[1].meta)
in1 = ff < inputs.input_file('input.mp4', v1, a1)
in2 = ff < inputs.input_file('input.mp4', v2, a2)
p1 = in1.video | filters.Trim(VIDEO, 2.0, 3.0) | filters.SetPTS(VIDEO)
p2 = in2.video | filters.Trim(VIDEO, 1.0, 2.0) | filters.SetPTS(VIDEO)
output = outputs.output_file('output.mp4',
codecs.VideoCodec('libx264'))
concat = p1 | filters.Concat(VIDEO)
p2 | concat > output
ff > output
ff.check_buffering()
def transcode(self) -> None:
""" Transcodes video
* checks source mediainfo
* runs `ffmpeg`
* validates result
"""
audio_meta, video_meta = self.get_meta_data(self.source)
# Get source mediainfo to use in validation
source_media_info = self.get_media_info(video_meta, audio_meta)
# set group of pixels length to segment size
gop = math.floor(source_media_info[VIDEO_FRAME_RATE] * GOP_DURATION)
# preserve original video FPS
vrate = source_media_info[VIDEO_FRAME_RATE]
# preserve source audio sampling rate
arate = source_media_info[AUDIO_SAMPLING_RATE]
# Common ffmpeg flags
ff = FFMPEG(overwrite=True, loglevel='repeat+level+info')
# Init source file
ff < input_file(self.source, Stream(VIDEO, video_meta),
Stream(AUDIO, audio_meta))
# Output codecs
video_opts = cast(Dict[str, Any], TRANSCODING_OPTIONS[VIDEO_CODEC])
cv0 = VideoCodec(gop=gop, rate=vrate, **video_opts)
audio_opts = cast(Dict[str, Any], TRANSCODING_OPTIONS[AUDIO_CODEC])
ca0 = AudioCodec(rate=arate, **audio_opts)
# Scaling
ff.video | Scale(**TRANSCODING_OPTIONS[SCALE]) > cv0
# codecs, muxer and output path
ff > output_file(self.destination, cv0, ca0, format='mp4')
# Run ffmpeg
self.run(ff)
# Get result mediainfo
audio_meta, video_meta = self.get_meta_data(self.destination)
dest_media_info = self.get_media_info(video_meta, audio_meta)
# Validate ffmpeg result
self.validate(source_media_info, dest_media_info)
def test_concat_audio_metadata(self):
"""
Concat filter sums samples count for audio streams.
"""
audio_meta = audio_meta_data(duration=1000.0,
sampling_rate=24000,
samples_count=24000 * 1000)
a = inputs.Stream(AUDIO, meta=audio_meta)
self.input_list.append(inputs.input_file('second.mp4', a))
concat = a | Concat(AUDIO)
self.source | concat
concat > self.output
am = cast(AudioMeta, self.output.codecs[-1].get_meta_data())
self.assertEqual(self.audio_metadata.duration + audio_meta.duration,
am.duration)
self.assertEqual(round(am.duration * audio_meta.sampling_rate),
am.samples)
def test_overlay_metadata(self):
"""
overlay takes bottom stream metadata
$ ffmpeg -y -i source.mp4 -i logo.mp4 -t 1 \
-filter_complex '[0:v][1:v]overlay=x=100:y=100' test.mp4
"""
vs = inputs.Stream(VIDEO, meta=video_meta_data(width=100, height=100))
self.input_list.append(inputs.input_file('logo.png', vs))
overlay = self.source | Overlay(x=self.video_metadata.width - 2,
y=self.video_metadata.height - 2)
vs | overlay
overlay > self.output
expected = '[0:v][1:v]overlay=x=1918:y=1078[vout0]'
self.assertEqual(expected, self.fc.render())
vm = cast(VideoMeta, self.output.codecs[0].get_meta_data())
self.assertEqual(vm.width, self.video_metadata.width)
self.assertEqual(vm.height, self.video_metadata.height)
def test_concat_video_metadata(self):
"""
Concat filter sums stream duration
$ ffmpeg -y -i first.mp4 -i second.mp4 -filter_complex concat test.mp4
"""
video_meta = video_meta_data(duration=1000.0,
frame_count=10000,
frame_rate=10.0)
vs = inputs.Stream(VIDEO, meta=video_meta)
self.input_list.append(inputs.input_file('second.mp4', vs))
concat = vs | Concat(VIDEO)
self.source | concat
concat > self.output
vm = cast(VideoMeta, self.output.codecs[0].get_meta_data())
self.assertEqual(self.video_metadata.duration + vs.meta.duration,
vm.duration)
self.assertEqual(self.video_metadata.frames + video_meta.frames,
vm.frames)
def test_codec_metadata_transform(self):
"""
Codecs parameters applied to stream metadata when using transform.
"""
with self.subTest('codec with transform'):
self.source.audio > self.output
am = cast(AudioMeta, self.output.codecs[1].meta)
self.assertEqual(am.bitrate, self.target_audio_bitrate)
with self.subTest('no input metadata'):
no_meta_input = inputs.input_file('input.mp4')
output = outputs.output_file('output.mp4',
codecs.AudioCodec('aac'))
no_meta_input.audio > output.audio
self.assertIsNone(output.codecs[0].meta)
with self.subTest('no transform'):
output = outputs.output_file('output.mp4',
codecs.AudioCodec('aac'))
self.source.audio > output.audio
am = cast(AudioMeta, output.codecs[0].meta)
self.assertEqual(am.bitrate, self.audio_metadata.bitrate)
def test_detect_trim_buffering(self):
"""
When trim and concat filters are used for editing timeline, buffering
may occur if order of scenes in output file does not match order of same
scenes in input file.
"""
cases = [
(False, [1.0, 2.0], [2.0, 3.0]),
(True, [2.0, 3.0], [1.0, 2.0]),
(True, [2.0, 3.0], [2.0, 4.0]),
]
for case in cases:
with self.subTest(case):
raises, first, second = case
ff = FFMPEG()
s1 = inputs.Stream(VIDEO, self.source.streams[0].meta)
s2 = inputs.Stream(VIDEO, self.source.streams[1].meta)
ff < inputs.input_file('input.mp4', s1, s2)
split = ff.video | filters.Split(VIDEO)
t1 = split | filters.Trim(VIDEO, *first)
p1 = t1 | filters.SetPTS(VIDEO)
t2 = split | filters.Trim(VIDEO, *second)
p2 = t2 | filters.SetPTS(VIDEO)
concat = p1 | filters.Concat(VIDEO)
output = outputs.output_file('output.mp4',
codecs.VideoCodec('libx264'))
p2 | concat > output
ff > output
try:
ff.check_buffering()
except BufferError as e:
self.assertTrue(raises, e)
else:
self.assertFalse(raises)
def test_filter_graph(self):
""" Filter complex smoke test and features demo.
[I-1/Logo]---<Scale>-------
|
[I-0/input]--<Deint>--<Overlay>--<Split>--<Scale>--[O/480p]
|
------<Scale>--[O/720p]
"""
vs = inputs.Stream(VIDEO)
logo = inputs.input_file('logo.png', vs)
self.input_list.append(logo)
out1 = outputs.output_file('out1.mp4')
self.output_list.append(out1)
deint = Deint()
deint.enabled = False # deinterlace is skipped
# first video stream is deinterlaced
next_node = self.source | deint
left, top = 20, 20 # logo position
# first overlay input is deinterlaced source (or source itself as
# deint filter is disabled)
over = next_node | Overlay(left, top)
logo_width, logo_height = 200, 50 # logo scaled
# second input stream is connected to logo scaler, followed by overlay
# filter
next_node = vs | Scale(logo_width, logo_height) | over
# audio is split to two streams
asplit = self.source | Split(AUDIO)
for out in self.output_list:
asplit > out
# video split to two steams
# connect split filter to overlayed video stream
split = next_node | Split(VIDEO)
# intermediate video stream scaling
sizes = [(640, 480), (1280, 720)]
for out, size in zip(self.output_list, sizes):
# add scale filters to video streams
w, h = size
scale = Scale(w, h)
# connect scaled streams to video destinations
split | scale > out
result = self.fc.render()
expected = ';'.join([
# overlay logo
'[0:v][v:scale0]overlay=x=20:y=20[v:overlay0]',
# split video to two streams
'[v:overlay0]split[v:split0][v:split1]',
# each video is scaled to own size
'[v:split0]scale=w=640:h=480[vout0]',
'[v:split1]scale=w=1280:h=720[vout1]',
# split audio to two streams
'[0:a]asplit[aout0][aout1]',
# logo scaling
'[1:v]scale=w=200:h=50[v:scale0]',
])
self.assertEqual(expected.replace(';', ';\n'),
result.replace(';', ';\n'))
def fc_factory():
src = inputs.input_file("input.mp4", inputs.Stream(VIDEO))
dst = outputs.output_file('output.mp4')
fc = FilterComplex(inputs.InputList((src, )),
outputs.OutputList((dst, )))
return fc, src, dst
