Video Data Loading

Load video data for curation using NeMo Curator.

How it Works

NeMo Curator loads videos with a composite stage that discovers files and extracts metadata:

1. VideoReader decomposes into a partitioning stage plus a reader stage.
2. Local paths use FilePartitioningStage to list files; remote URLs (for example, s3://, gcs://, http(s)://) use ClientPartitioningStage backed by fsspec.
3. For remote datasets, you can optionally supply an explicit file list using ClientPartitioningStage.input_list_json_path.
4. VideoReaderStage downloads bytes (local or via FSPath) and calls video.populate_metadata() to extract resolution, fps, duration, encoding format, and other fields.
5. Set video_limit to cap discovery; use None for unlimited. Set verbose=True to log detailed per-video information.

Local and Cloud

Use VideoReader to load videos from local paths or remote URLs.

Local Paths

• Examples: /data/videos/, /mnt/datasets/av/
• Uses FilePartitioningStage to recursively discover files.
• Filters by extensions: .mp4, .mov, .avi, .mkv, .webm.
• Set video_limit to cap discovery during testing (None means unlimited).

Remote Paths

• Examples: s3://bucket/path/, gcs://bucket/path/, https://host/path/, and other fsspec-supported protocols such as s3a:// and abfs://.
• Uses ClientPartitioningStage backed by fsspec to list files.
• Optional input_list_json_path allows explicit file lists under a root prefix.
• Wraps entries as FSPath for efficient byte access during reading.

Example

1
from nemo_curator.pipeline import Pipeline
2
from nemo_curator.stages.video.io.video_reader import VideoReader
3
4
pipe = Pipeline(name="video_read", description="Read videos and extract metadata")
5
pipe.add_stage(VideoReader(input_video_path="s3://my-bucket/videos/", video_limit=None, verbose=True))
6
pipe.run()

Explicit File List (JSON)

For remote datasets, ClientPartitioningStage can use an explicit file list JSON. Each entry must be an absolute path under the specified root.

JSON Format

1
[
2
  "s3://my-bucket/datasets/videos/video1.mp4",
3
  "s3://my-bucket/datasets/videos/video2.mkv",
4
  "s3://my-bucket/datasets/more_videos/video3.webm"
5
]

If any entry is outside the root, the stage raises an error.

Example

1
from nemo_curator.pipeline import Pipeline
2
from nemo_curator.stages.client_partitioning import ClientPartitioningStage
3
from nemo_curator.stages.video.io.video_reader import VideoReaderStage
4
5
ROOT = "s3://my-bucket/datasets/"
6
JSON_LIST = "s3://my-bucket/lists/videos.json"
7
8
pipe = Pipeline(name="video_read_json_list", description="Read specific videos via JSON list")
9
pipe.add_stage(
10
    ClientPartitioningStage(
11
        file_paths=ROOT,
12
        input_list_json_path=JSON_LIST,
13
        files_per_partition=1,
14
        file_extensions=[".mp4", ".mov", ".avi", ".mkv", ".webm"],
15
    )
16
)
17
pipe.add_stage(VideoReaderStage(verbose=True))
18
pipe.run()

Supported File Types

The loader filters these video extensions by default:

• .mp4
• .mov
• .avi
• .mkv
• .webm

Metadata on Load

After a successful read, the loader populates the following metadata fields for each video:

• size (bytes)
• width, height
• framerate
• num_frames
• duration (seconds)
• video_codec, pixel_format, audio_codec
• bit_rate_k