sk-video

What it is and what it does

This package provides video I/O and analysis capabilities, wrapping FFmpeg to read and write video frames, extract metadata, and compute video quality metrics. It follows the conventions of scikit-image and scikit-learn, sitting on top of numpy and scipy for workflows already using the scientific Python stack.

The library provides frame-by-frame video access, metadata inspection, and implementations of standard quality metrics (MSE, PSNR, SSIM, MS-SSIM, NIQE, BRISQUE, VIIDEO, Video-BLIINDS, ST-RRED). Recent versions validated metric implementations against reference code and fixed divergences; the 1.1.12+ line added pathlib.Path support, URL and file-like object I/O, and audio passthrough. The project is actively maintained with ongoing compatibility work for modern NumPy and Python versions.

Use it for:

• Extract frames from video files and process them with numpy and scipy for computer vision tasks.
• Compute video quality metrics (SSIM, PSNR, NIQE, BRISQUE) for compression or enhancement evaluation.
• Read video metadata and properties without decoding all frames.
• Batch process video files to generate frame sequences for machine learning pipelines.
• Validate video codec compatibility and inspect FFmpeg-supported container formats.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Reads, writes, and processes video files using FFmpeg as its backend, providing video I/O and quality metrics (NIQE, BRISQUE, SSIM, PSNR) built on numpy and scipy.

Yes, if you need video I/O and quality metrics in a Python environment already using numpy and scipy. Install friction is low and maintenance is active. Requires FFmpeg >= 2.8 on system PATH and Python >= 3.10. Metric implementations changed in 1.2.0 (more accurate but incomparable to 1.1.x scores); verify this does not break existing workflows. No known security vulnerabilities.

Install

sk-video on PyPI

pip install sk-video

uv add sk-video

poetry add sk-video

Installing sk-video

Before you install

Low install friction; pure Python wheel with only numpy and scipy as runtime dependencies. Actively maintained with recent metric accuracy fixes and NumPy 2.x compatibility. Requires FFmpeg >= 2.8 on system PATH and Python >= 3.10.

License in practice

BSD permissive license allows commercial and private use with minimal restrictions; attribution and license notice required in distributions.

Quickstart

pip install scikit-video

import skvideo.io
video_data = skvideo.io.vread('video.mp4')
metadata = skvideo.io.ffprobe('video.mp4')

FFmpeg >= 2.8 must be installed and on system PATH; Python >= 3.10 required.

Verify before relying

• Whether the metric accuracy changes in 1.2.0 (NIQE, BRISQUE, VIIDEO, Video-BLIINDS) affect existing workflows or benchmarks.
• Current state of the deprecated libav/avconv backend and mediainfo support (planned removal in 1.4.0).
• Whether Pillow >= 9.0 is a runtime or optional dependency.

Package facts

Evidence: sk_video-1.1.10-py2.py3-none-any.whl

Tags

More Scientific/Engineering packages

Further reading