def test_spectrogram_image(spark: SparkSession):
"""Test generate spectrogram image
from YouTubeVideo/VideoStream videos types."""
videostream_df = spark.createDataFrame(
[
(VideoStream(uri=os.path.abspath(
os.path.join(
os.path.dirname(__file__),
"..",
"assets",
"big_buck_bunny_short.mp4",
))), ),
],
["video"],
)
youtube_df = spark.createDataFrame(
[
(YouTubeVideo(vid="rUWxSEwctFU"), ),
],
["video"],
)
videostream_df = videostream_df.withColumn(
"spectrogram",
spectrogram_image(col("video")),
)
youtube_df = youtube_df.withColumn(
"spectrogram",
spectrogram_image(col("video")),
)
videostream_sample = videostream_df.first()["spectrogram"]
youtube_sample = youtube_df.first()["spectrogram"]
assert type(videostream_sample) == Image
assert type(youtube_sample) == Image
def test_spectrogram_image(
spark: SparkSession, tmp_path: Path, asset_path: Path
):
"""Test generate spectrogram image
from YouTubeVideo/VideoStream videos types."""
video = VideoStream(str(asset_path / "big_buck_bunny_short.mp4"))
s1 = (
spark.createDataFrame([(video,)], ["video"])
.withColumn(
"spectrogram",
spectrogram_image(col("video"), lit(str(tmp_path / "s1.jpg"))),
)
.first()["spectrogram"]
)
assert type(s1) == Image
yt = YouTubeVideo(vid="rUWxSEwctFU")
s2 = (
spark.createDataFrame([(yt,)], ["video"])
.withColumn(
"spectrogram",
spectrogram_image(col("video"), lit(str(tmp_path / "s2.jpg"))),
)
.first()["spectrogram"]
)
assert type(s2) == Image
def test_video_to_images(
spark: SparkSession, tmp_path: Path, asset_path: Path
):
"""Test extract video frames from YouTubeVideo/VideoStream types
into list of Image assets.
"""
sample_rate = 2
max_samples = 10
video = VideoStream(str(asset_path / "big_buck_bunny_short.mp4"))
df1 = spark.createDataFrame(
[(video, Segment(0, 20))], ["video", "segment"]
)
output_dir = tmp_path / "videostream_test"
output_dir.mkdir(parents=True)
df1 = df1.withColumn(
"images",
video_to_images(
col("video"),
lit(str(output_dir)),
col("segment"),
lit(sample_rate),
lit(max_samples),
),
)
df2 = spark.createDataFrame(
[(YouTubeVideo(vid="rUWxSEwctFU"), Segment(0, 20))],
["video", "segment"],
)
output_dir = tmp_path / "youtube_test"
output_dir.mkdir(parents=True)
df2 = df2.withColumn(
"images",
video_to_images(
col("video"),
lit(str(output_dir)),
col("segment"),
lit(sample_rate),
lit(max_samples),
),
)
videostream_sample = df1.first()["images"]
youtube_sample = df2.first()["images"]
assert (
type(videostream_sample) == list
and type(videostream_sample[0]) == Image
and len(videostream_sample) == max_samples
)
assert (
type(youtube_sample) == list
and type(youtube_sample[0]) == Image
and len(youtube_sample) == max_samples
)
def test_video_to_images(spark: SparkSession):
"""Test extract video frames from YouTubeVideo/VideoStream types
into list of Image assets.
"""
sample_rate = 2
max_samples = 10
videostream_df = spark.createDataFrame(
[
(
VideoStream(uri=os.path.abspath(
os.path.join(
os.path.dirname(__file__),
"..",
"assets",
"big_buck_bunny_short.mp4",
))),
Segment(0, 20),
),
],
["video", "segment"],
)
youtube_df = spark.createDataFrame(
[
(YouTubeVideo(vid="rUWxSEwctFU"), Segment(0, 20)),
],
["video", "segment"],
)
videostream_df = videostream_df.withColumn(
"images",
video_to_images(col("video"), col("segment"), lit(sample_rate),
lit(max_samples)),
)
youtube_df = youtube_df.withColumn(
"images",
video_to_images(col("video"), col("segment"), lit(sample_rate),
lit(max_samples)),
)
videostream_sample = videostream_df.first()["images"]
youtube_sample = youtube_df.first()["images"]
assert (type(videostream_sample) == list
and type(videostream_sample[0]) == Image
and len(videostream_sample) == max_samples)
assert (type(youtube_sample) == list and type(youtube_sample[0]) == Image
and len(youtube_sample) == max_samples)
def test_scene_detect(spark: SparkSession, asset_path: Path):
video = VideoStream(str(asset_path / "big_buck_bunny_short.mp4"))
df = spark.createDataFrame([(video, )], ["video"])
result = [
r.asDict(True) for r in df.withColumn("scenes", scene_detect(
"video")).first()["scenes"]
]
expected = [{
"start": {
"frame_num": 0,
"frame_pos_sec": 0.0
},
"end": {
"frame_num": 300,
"frame_pos_sec": 10.010000228881836
},
}]
for rs, xp in zip(result, expected):
pdt.assert_frame_equal(pd.DataFrame(rs), pd.DataFrame(xp))
def test_video_metadata(spark: SparkSession, asset_path: Path):
video = VideoStream(str(asset_path / "big_buck_bunny_short.mp4"))
result = (spark.createDataFrame([(video, )], ["video"]).select(
video_metadata(col("video")).alias("meta")).first()["meta"].asDict())
expected = {
"width": 640,
"height": 360,
"num_frames": 300,
"duration": 10.010000228881836,
"bit_rate": 415543,
"frame_rate": 30,
"codec": "h264",
"size": 736613,
"_errors": None,
}
pdt.assert_series_equal(pd.Series(result), pd.Series(expected))
video = "bad_uri"
result = (spark.createDataFrame([(video, )], ["video"]).select(
video_metadata(col("video")).alias("meta")).first()["meta"].asDict())
err = result["_errors"].asDict()
assert err["message"].startswith("ffprobe error")
assert "bad_uri: No such file or directory" in err["stderr"]
def test_videostream(spark, tmpdir):
df = spark.createDataFrame(
[Row(VideoStream("uri1")),
Row(VideoStream("uri2"))])
_check_roundtrip(spark, df, tmpdir)
def deserialize(self, datum) -> "VideoStream":
from rikai.types import VideoStream # pylint: disable=import-outside-toplevel
return VideoStream(datum[0])
def test_videostream(self):
df = self.spark.createDataFrame(
[Row(VideoStream("uri1")), Row(VideoStream("uri2"))]
)
self._check_roundtrip(df)