/keyframe-detection

Primary LanguagePython

Keyframe Detection

keyframe is a Python package designed for detecting keyframes in videos using various algorithms. This package provides an easy-to-use interface for processing video frames and extracting significant frames (keyframes) based on different detection methods.

Available Algorithms

  • pixelwise: Compares pixels between frames.
  • background_subtraction: Uses background subtraction to detect significant changes.
  • optical_flow: Measures the movement between frames using optical flow.
  • histogram_comparison: Compares histograms of frames.
  • sift_matching: Uses SIFT features to detect significant changes.
  • mean_squared_error: Computes the mean squared error between frames.
  • phase_correlation: Uses phase correlation to detect keyframes.

Installation

To install the package, use pip:

pip install keyframe

Usage

import os
from keyframe.framer import Framer

framer = Framer()

video_path = "asset/short.mp4"
algorithm = "pixelwise"  # [pixelwise, background_subtraction, optical_flow, histogram_comparison, sift_matching, mean_squared_error, phase_correlation]
output_dir = f"output/{algorithm}_keyframes"

if not os.path.exists(output_dir):
    os.makedirs(output_dir)

print(f"Processing video with '{algorithm}' algorithm...")
framer.process_video(video_path, algorithm, output_dir)

Command-Line Interface (CLI)

You can use the CLI to detect keyframes in a video. The detected keyframes will be saved to a specified directory.

python cli.py path/to/video.mp4 --algorithm pixelwise --output-dir keyframes_output
  • video_path: Path to the video file.
  • --algorithm: The algorithm to use for keyframe detection (default: mean_squared_error).
  • --output-dir: Directory to save the detected keyframes (default: keyframes).

Example

python cli.py example_video.mp4 --algorithm optical_flow --output-dir keyframes_output

This command processes example_video.mp4 using the optical_flow algorithm and saves the keyframes in the keyframes_output directory.

Contributing

Contributions are welcome! Please open an issue or submit a pull request if you have suggestions or improvements.

License

This project is licensed under the MIT License - see the LICENSE file for details.