TileDB-Video-Py

TileDB-Video-Py is a Python library for writing a video stream as a TileDB array and slicing it efficiently.

Installation

This project is currently in early development and can be installed directly from GitHub:

pip install git+https://github.com/TileDB-Inc/TileDB-Video-Py

Usage

To see tiledb.video in action, you may view or run locally the Jupyter notebook demo.

API

The API consists of the following functions under the tiledb.video namespace package:

Writing to TileDB

• from_file(uri, file, *, split_duration=1.0, start_time=None, end_time=None, format="mp4", stream_index=0)

  Create a TileDB array at given URI from a video file stream.

  • uri: URI for new TileDB array
  • file: Input video file path or file-like object
  • split_duration: Target duration of each split video segment (in seconds)
  • start_time: Start time offset (in seconds)
  • end_time: End time offset (in seconds)
  • format: Format of the split video segments
  • stream_index: Index of the stream channel to write

Reading from TileDB

• get_codec_context(uri)

  Get a subset of the video stream codec context attributes.

  • uri: URI of TileDB array to read from

• to_file(uri, file, *, format="mp4", start_time=None, end_time=None)

  Read a video from a TileDB array into a file.

  • uri: URI of TileDB array to read from
  • file: Output video file path or file-like object
  • start_time: Start time offset (in seconds)
  • end_time: End time offset (in seconds)
  • format: Format of the output video file

• iter_images(uri, *, start_time=None, end_time=None, max_threads=1, width=None, height=None, src_colorspace=None, dst_colorspace=None, interpolation=None)

  Return an iterator of RGB images represented as PIL.Image instances from a TileDB array.

  • uri: URI of TileDB array to read from
  • start_time: Start time offset (in seconds)
  • end_time: End time offset (in seconds)
  • max_threads: Max number of threads to fetch segments concurrently, or 0 to fetch all segments in a single call

  For the remaining optional parameters, please consult the documentation of the PyAV VideoReformatter.reformat() method.

• iter_ndarrays(uri, *, start_time=None, end_time=None, max_threads=1, format="rgb24", width=None, height=None, src_colorspace=None, dst_colorspace=None, interpolation=None)

  Return an iterator of images represented as Numpy arrays from a TileDB array.

  • uri: URI of TileDB array to read from
  • start_time: Start time offset (in seconds)
  • end_time: End time offset (in seconds)
  • max_threads: Max number of threads to fetch segments concurrently, or 0 to fetch all segments in a single call

  For the remaining optional parameters, please consult the documentation of the PyAV VideoReformatter.reformat() method.

CLI

A simple tiledb-video command line interface is also provided for converting between video files and TileDB arrays:

$ tiledb-video --help
usage: tiledb-video [-h] [--from FROM] [--to TO] [-f FORMAT]
                    [-l SEGMENT_LENGTH] [-s STREAM]
                    input output

Convert a video file to or from TileDB

positional arguments:
  input                 Input video file path or TileDB URI
  output                Output video file path or TileDB URI

optional arguments:
  -h, --help            show this help message and exit
  --from FROM           Start time offset (in seconds) (default: None)
  --to TO               End time offset (in seconds) (default: None)
  -f FORMAT, --format FORMAT
                        Format of the output video file (for conversion from
                        tileDB) or the split video segments (for conversion to
                        tileDB) (default: mp4)

Arguments for conversion to TileDB:
  -l SEGMENT_LENGTH, --segment-length SEGMENT_LENGTH
                        Split video segment length (in seconds) (default: 5.0)
  -s STREAM, --stream STREAM
                        Index of the stream channel to read (default: 0)