Desplice

Desplice is a Python library designed to process video files by removing duplicate frames, still images, and certain transitions. This helps in deduplicating video content, making it more efficient for storage or further processing.

Features

• Heal: Drops duplicate frames from the output.
• Keep: Retains one instance of each duplicate frame in the output.
• Explode: Removes duplicate frames from the video output and returns separate clips for each splice.

Installation

To install Desplice, you can use pip:

pip install desplice

Usage

Below are some examples demonstrating how to use Desplice.

Importing the Library

from desplice import Desplice

Loading a Video and Processing it

Take note that in the input value may be a path to the target video or alternatively a list of frames list[np.ndarray] that has already been captured/loaded.

# Initialize the Desplice class
desplice = Desplice(debug=True)

# Load a video file into memory
video_frames = desplice.load_video_to_memory('path_to_your_video.mp4')

# Process the video to deduplicate frames
result, is_slideshow = desplice.process(video_frames, mode='heal', show=True, show_breaks=True)

If show is set to True processing and results will be displayed using cv2.imshow().

If show_breaks is set to True one black frame will be inserted between each slice.

If debug is set to True measurements and detections will be returned to console.

Example of Different Modes:

Heal Mode

Removes the detected content from output entirely.

result, is_slideshow = desplice.process('path_to_your_video.mp4', mode='heal')

Keep Mode

Keeps 1 frame of the duplicated content in the output.

result, is_slideshow = desplice.process('path_to_your_video.mp4', mode='keep')

Explode Mode

Returns each segment inbetween each duplicated sequence as it's own sequence.

result, is_slideshow = desplice.process('path_to_your_video.mp4', mode='explode')

Results are returned in a nested tuple object:

(
    (
        [video-1, video-2, video-3...],  # List of videos: list[list[np.ndarray]]
        [still-1, still-2, still-3]  # List of still images: list[np.ndarray]
    ), 
    slideshow  # Is the video a slideshow: Bool
)

Deduplication weights can be passed to augment the behavior of the antidupe module:

custom_thresholds = {
        'ih': 0.0,  # Image Hash
        'ssim': 0.1,  # SSIM
        'cs': 0.02,  # Cosine Similarity
        'cnn': 0.03,  # CNN
        'dedup': 0.005  # Mobilenet
    }

desplice = Desplice(thresholds=custom_thresholds)

Further information on using antidupe can be found here

Class and Method Descriptions

Desplice

Methods

• __init__(self, thresholds: dict = None, device: str = 'cpu', debug: bool = False): Initializes the Desplice class with optional threshold values, device, and debug mode.

• load_video_to_memory(file_path: str) -> list: Loads a video from the given file path into memory.

• show_video_from_frames(frames: list): Displays a video from a list of frames.

• deduplicate_frames(self, frames: list) -> tuple: Deduplicates frames from the input video and returns a tuple containing processed video frames and images.

• process(self, file_path_or_array: [str, list[np.ndarray]], mode: modes = 'heal', show: bool = False, show_breaks: bool = False) -> tuple: Processes the video based on the specified mode and returns the processed result and a boolean indicating if the output is a slideshow.

License

This project is licensed under the Apache License, Version 2.0. See the LICENSE file for more details.