scikit-video: Video Processing in Python

Warnings

• breaking Incompatibility with recent NumPy versions (>=1.20, especially >=1.24) due to the removal of `numpy.float` and `numpy.int` aliases, causing `AttributeError`.
  Fix: Downgrade NumPy to a version prior to 1.20 (e.g., `pip install numpy==1.19.5`) or apply a monkey-patch before `skvideo` imports: `import numpy; numpy.float = numpy.float64; numpy.int = numpy.int_`.
• gotcha Older installations of `scikit-video` might have used the package name `sk-video`, leading to import conflicts or unexpected behavior if `scikit-video` is installed afterwards.
  Fix: Ensure previous installations of `sk-video` are uninstalled before installing `scikit-video`: `pip uninstall sk-video` then `pip install scikit-video`.
• gotcha The `skvideo.io.vread` function loads the entire video into memory. For large video files, this can lead to `MemoryError` or system instability.
  Fix: For large videos, use `skvideo.io.vreader` which returns a generator, allowing frame-by-frame processing and significantly reducing memory footprint.
• gotcha `skvideo.io.vread` can be significantly slower if `num_frames` is not explicitly provided, especially for certain video types, as it may perform multiple passes to auto-detect video properties.
  Fix: Always specify `num_frames` when reading only a subset of frames (e.g., `skvideo.io.vread(filename, num_frames=N)`).
• gotcha Requires external `ffmpeg` or `libav` binaries to be installed and available in the system's PATH. Without them, video I/O operations will fail.
  Fix: Install `ffmpeg` (recommended) or `libav` on your system. If they are not in PATH, you can explicitly set their location using `skvideo.setFFmpegPath('/path/to/ffmpeg')` or `skvideo.setLibAVPath('/path/to/libav')`.
• deprecated The `setup.py` script, used for installing from source, relies on `numpy.distutils`, which is deprecated and removed in Python 3.12 and later, potentially breaking source installations.
  Fix: For Python 3.12+, installation from source may require manual intervention or the use of `pip install --no-build-isolation .` or equivalent methods, if the project is not updated to `pyproject.toml`.

Install

• pip install scikit-video Install via pip

Imports

• skvideo.io

  import skvideo.io

  For video reading and writing functionalities.
• skvideo.datasets

  import skvideo.datasets

  Provides access to test datasets.
• skvideo.utils

  import skvideo.utils

  For various utility functions.

Quickstart

This quickstart demonstrates how to load a sample video from `skvideo.datasets` into a NumPy array using `skvideo.io.vread`. It also shows how to inspect the video's shape and data type. For large videos, `skvideo.io.vreader` is recommended for memory efficiency.

import skvideo.io
import skvideo.datasets
import numpy as np

# Get a sample video filename
filename = skvideo.datasets.bigbuckbunny()

# Read the video into a NumPy array
# For larger videos, consider skvideo.io.vreader for frame-by-frame processing
videodata = skvideo.io.vread(filename, num_frames=10) # Read first 10 frames for quick test

print(f"Video data shape: {videodata.shape}")
print(f"Data type: {videodata.dtype}")

# Example: Convert to grayscale (if not already)
if videodata.shape[-1] == 3:
    # Simple average for grayscale, or use more advanced conversion
    grayscale_video = np.mean(videodata, axis=-1, keepdims=True).astype(videodata.dtype)
    print(f"Grayscale video shape: {grayscale_video.shape}")

# To prevent 'module numpy has no attribute float' errors with recent NumPy, 
# you might need this workaround before importing anything that uses it if it breaks:
# import numpy
# numpy.float = numpy.float64
# numpy.int = numpy.int_