fastvideo.pipelines.preprocess.preprocess_pipeline_ode_trajectory

ODE Trajectory Data Preprocessing pipeline implementation.

This module contains an implementation of the ODE Trajectory Data Preprocessing pipeline using the modular pipeline architecture.

Sec 4.3 of CausVid paper: https://arxiv.org/pdf/2412.07772

Module Contents

Classes

PreprocessPipeline_ODE_Trajectory

ODE Trajectory preprocessing pipeline implementation.

Data

EntryClass
logger

API

fastvideo.pipelines.preprocess.preprocess_pipeline_ode_trajectory.EntryClass

None

class fastvideo.pipelines.preprocess.preprocess_pipeline_ode_trajectory.PreprocessPipeline_ODE_Trajectory(model_path: str, fastvideo_args: fastvideo.fastvideo_args.FastVideoArgs | fastvideo.fastvideo_args.TrainingArgs, required_config_modules: list[str] | None = None, loaded_modules: dict[str, torch.nn.Module] | None = None)

Bases: fastvideo.pipelines.preprocess.preprocess_pipeline_base.BasePreprocessPipeline

ODE Trajectory preprocessing pipeline implementation.

Initialization

Initialize the pipeline. After init, the pipeline should be ready to use. The pipeline should be stateless and not hold any batch state.

create_pipeline_stages(fastvideo_args: fastvideo.fastvideo_args.FastVideoArgs)

Set up pipeline stages with proper dependency injection.

forward(batch: fastvideo.pipelines.pipeline_batch_info.ForwardBatch, fastvideo_args: fastvideo.fastvideo_args.FastVideoArgs, args)

get_pyarrow_schema() → pyarrow.Schema

Return the PyArrow schema for ODE Trajectory pipeline.

num_processed_samples: int

None

pbar: Any

None

preprocess_dataloader: torchdata.stateful_dataloader.StatefulDataLoader

None

preprocess_loader_iter: collections.abc.Iterator[dict[str, Any]]

None

preprocess_text_and_trajectory(fastvideo_args: fastvideo.fastvideo_args.FastVideoArgs, args)

Preprocess text-only data and generate trajectory information.

fastvideo.pipelines.preprocess.preprocess_pipeline_ode_trajectory.logger

‘init_logger(…)’