nemo_curator.stages.video.clipping.clip_extraction_stages

Module Contents

Classes

Name	Description
`ClipTranscodingStage`	Stage that transcodes video clips into a standardized format.
`FixedStrideExtractorStage`	Stage that extracts video clips using fixed-length intervals.

API

class nemo_curator.stages.video.clipping.clip_extraction_stages.ClipTranscodingStage(
    num_cpus_per_worker: float = 6.0,
    encoder: str = 'libx264',
    encoder_threads: int = 1,
    encode_batch_size: int = 16,
    nb_streams_per_gpu: int = 3,
    use_hwaccel: bool = False,
    use_input_bit_rate: bool = False,
    num_clips_per_chunk: int = 32,
    ffmpeg_verbose: bool = False,
    verbose: bool = False,
    name: str = 'clip_transcoding'
)

Dataclass

Bases: ProcessingStage[VideoTask, VideoTask]

Stage that transcodes video clips into a standardized format.

This stage handles the conversion of video clips using FFmpeg, supporting both software (libx264, libopenh264) and hardware (NVENC) encoding with configurable parameters.

Parameters:

num_cpus_per_worker

floatDefaults to 6.0

Number of CPUs per worker.

encoder

strDefaults to 'libx264'

Video encoder to use.

encoder_threads

intDefaults to 1

Number of threads per encoder.

encode_batch_size

intDefaults to 16

Number of clips to encode in parallel.

nb_streams_per_gpu

intDefaults to 3

Number of streams per GPU.

use_hwaccel

boolDefaults to False

Whether to use hardware acceleration.

use_input_bit_rate

boolDefaults to False

Whether to use input video bit rate.

num_clips_per_chunk

intDefaults to 32

Number of clips per chunk. If the number of clips is larger than this, the clips will be split into chunks, and created VideoTasks for each chunk.

verbose

boolDefaults to False

Whether to print verbose logs.

ffmpeg_verbose

boolDefaults to False

Whether to print FFmpeg verbose logs.

encode_batch_size

int = 16

encoder

str = 'libx264'

encoder_threads

int = 1

ffmpeg_verbose

bool = False

name

str = 'clip_transcoding'

nb_streams_per_gpu

int = 3

num_clips_per_chunk

int = 32

num_cpus_per_worker

float = 6.0

use_hwaccel

bool = False

use_input_bit_rate

bool = False

verbose

bool = False

nemo_curator.stages.video.clipping.clip_extraction_stages.ClipTranscodingStage.__post_init__() -> None

Post-initialization method called after all fields are set.

nemo_curator.stages.video.clipping.clip_extraction_stages.ClipTranscodingStage._add_decoder_threads(
    command: list[str]
) -> None

Add decoder thread options to command.

nemo_curator.stages.video.clipping.clip_extraction_stages.ClipTranscodingStage._add_hwaccel_options(
    command: list[str]
) -> None

Add hardware acceleration options to command.

nemo_curator.stages.video.clipping.clip_extraction_stages.ClipTranscodingStage._add_input_options(
    command: list[str],
    clip: nemo_curator.tasks.video.Clip,
    video_filename: str,
    index: int
) -> None

Add input options to command.

nemo_curator.stages.video.clipping.clip_extraction_stages.ClipTranscodingStage._add_nvenc_options(
    command: list[str],
    force_pix_fmt: bool
) -> None

Add NVENC-specific encoding options.

nemo_curator.stages.video.clipping.clip_extraction_stages.ClipTranscodingStage._add_output_options(
    command: list[str],
    clip: nemo_curator.tasks.video.Clip,
    index: int
) -> None

Add output options to command.

nemo_curator.stages.video.clipping.clip_extraction_stages.ClipTranscodingStage._add_video_encoding_options(
    command: list[str],
    use_bit_rate: str | None,
    force_pix_fmt: bool
) -> None

Add video encoding options to command.

nemo_curator.stages.video.clipping.clip_extraction_stages.ClipTranscodingStage._build_ffmpeg_command(
    video_filename: str,
    clips: list[nemo_curator.tasks.video.Clip],
    force_pix_fmt: bool,
    use_bit_rate: str | None
) -> list[str]

Build the FFmpeg command for extracting clips.

nemo_curator.stages.video.clipping.clip_extraction_stages.ClipTranscodingStage._extract_clips(
    working_dir: pathlib.Path,
    video_filename: str,
    force_pix_fmt: bool,
    use_bit_rate: str | None,
    clips: list[nemo_curator.tasks.video.Clip]
) -> None

Extract clips using FFmpeg.

nemo_curator.stages.video.clipping.clip_extraction_stages.ClipTranscodingStage._handle_ffmpeg_error(
    error: subprocess.CalledProcessError,
    command: list[str],
    clips: list[nemo_curator.tasks.video.Clip]
) -> None

Handle FFmpeg command errors.

nemo_curator.stages.video.clipping.clip_extraction_stages.ClipTranscodingStage._read_clips_to_memory(
    working_dir: pathlib.Path,
    clips: list[nemo_curator.tasks.video.Clip]
) -> None

Read extracted clips back into memory.

nemo_curator.stages.video.clipping.clip_extraction_stages.ClipTranscodingStage._run_ffmpeg_command(
    command: list[str],
    working_dir: pathlib.Path,
    clips: list[nemo_curator.tasks.video.Clip]
) -> None

Run the FFmpeg command and handle errors.

nemo_curator.stages.video.clipping.clip_extraction_stages.ClipTranscodingStage.inputs() -> tuple[list[str], list[str]]

nemo_curator.stages.video.clipping.clip_extraction_stages.ClipTranscodingStage.outputs() -> tuple[list[str], list[str]]

nemo_curator.stages.video.clipping.clip_extraction_stages.ClipTranscodingStage.process(
    task: nemo_curator.tasks.video.VideoTask
) -> nemo_curator.tasks.video.VideoTask

nemo_curator.stages.video.clipping.clip_extraction_stages.ClipTranscodingStage.ray_stage_spec() -> dict[str, typing.Any]

Ray stage specification for this stage.

nemo_curator.stages.video.clipping.clip_extraction_stages.ClipTranscodingStage.setup(
    worker_metadata: nemo_curator.backends.base.WorkerMetadata | None = None
) -> None

Setup method called once before processing begins. Override this method to perform any initialization that should happen once per worker. Args: worker_metadata (WorkerMetadata, optional): Information about the worker (provided by some backends)

class nemo_curator.stages.video.clipping.clip_extraction_stages.FixedStrideExtractorStage(
    clip_len_s: float,
    clip_stride_s: float,
    min_clip_length_s: float,
    limit_clips: int,
    verbose: bool = False,
    name: str = 'fixed_stride_extractor'
)

Dataclass

Bases: ProcessingStage[VideoTask, VideoTask]

Stage that extracts video clips using fixed-length intervals.

This stage splits videos into clips of specified length and stride, ensuring each clip meets minimum length requirements and optionally limiting total clips.

clip_len_s

float

clip_stride_s

float

limit_clips

int

min_clip_length_s

float

name

str = 'fixed_stride_extractor'

verbose

bool = False

nemo_curator.stages.video.clipping.clip_extraction_stages.FixedStrideExtractorStage.inputs() -> tuple[list[str], list[str]]

nemo_curator.stages.video.clipping.clip_extraction_stages.FixedStrideExtractorStage.outputs() -> tuple[list[str], list[str]]

nemo_curator.stages.video.clipping.clip_extraction_stages.FixedStrideExtractorStage.process(
    task: nemo_curator.tasks.video.VideoTask
) -> nemo_curator.tasks.video.VideoTask