Implementation:NVIDIA NeMo Curator VideoReaderStage

Knowledge Sources	NeMo Curator
Domains	Data_Curation, Video_Processing
Last Updated	2026-02-14 17:00 GMT

Overview

Concrete tool for reading video files and extracting metadata provided by NeMo Curator.

Description

The VideoReaderStage reads video files from the local filesystem and extracts comprehensive metadata including dimensions, frame rate, duration, codecs, and other technical properties. It stores results in a VideoTask object containing the video source bytes and metadata.

Usage

Import this stage when building a video curation pipeline that needs to read raw video files from storage. Combine with FilePartitioningStage via the VideoReader composite stage for a complete ingestion solution.

Code Reference

Source Location

• Repository: NeMo Curator
• File: nemo_curator/stages/video/io/video_reader.py
• Lines: L79-290

Signature

@dataclass
class VideoReaderStage(ProcessingStage[FileGroupTask, VideoTask]):
    input_path: str | None = None
    verbose: bool = False
    name: str = "video_reader"

Import

from nemo_curator.stages.video.io.video_reader import VideoReaderStage

I/O Contract

Inputs

Name	Type	Required	Description
task	FileGroupTask	Yes	List of video file paths to read

Outputs

Name	Type	Description
task	VideoTask	Contains video.source_bytes and video.metadata (duration, fps, resolution, codec)

Usage Examples

from nemo_curator.stages.video.io.video_reader import VideoReader
from nemo_curator.pipeline import Pipeline

reader = VideoReader(
    input_video_path="./data/videos",
    video_limit=100,
    verbose=True,
)

pipeline = Pipeline()
pipeline.add_stage(reader)

Related Pages

Implements Principle

• Principle:NVIDIA_NeMo_Curator_Video_Ingestion
• Environment:NVIDIA_NeMo_Curator_Python_Linux_Base
• Environment:NVIDIA_NeMo_Curator_Video_Codec_Stack
• Environment:NVIDIA_NeMo_Curator_Ray_Cluster
• Heuristic:NVIDIA_NeMo_Curator_GPU_Memory_Resource_Allocation
• Heuristic:NVIDIA_NeMo_Curator_Video_Frame_Sampling_Strategy