cleanframes 0.3.11

A professional tool for cleaning duplicate or near-duplicate image frames using perceptual hashing and embeddings.

CleanFrames

CleanFrames is a Python library designed to clean and summarize image frames stored in folders efficiently using embedding models and clustering techniques. It processes folders of frames, removes duplicates or near-duplicates, caches embeddings and reports for faster subsequent runs, and saves cleaned/removed images alongside the original dataset.

Key Features

• Processes folders of image frames instead of videos.
• Supports multiple embedding models to represent frames.
• Various clustering methods to group similar frames.
• Caches embeddings and reports to optimize performance.
• Saves cleaned and removed images beside the original dataset.
• Visualization tools to inspect clusters and frame pairs.
• Generates text-only console reports summarizing cleaning results.

Installation

To install CleanFrames using pip:

pip install cleanframes

Usage

Basic Example

from cleanframes import CleanFrame

# Initialize with folder path, embedding model, clustering method, and caching enabled
cf = CleanFrame(path='path/to/frames_folder')

# Run the full cleaning pipeline: embedding, clustering, cleaning
cf.run()

# Generate a text-only console report of the cleaning results
cf.report()

# Visualize clusters of frames
cf.visualize_clusters()

Optimized Workflow Example

from cleanframes import CleanFrame

# Initialize with different model and clustering method
cf = CleanFrame(
    path='path/to/frames_folder',
    model='clip-ViT-L-14',
    cluster='dbscan',
    cache=True,
    verbose=True
)

# Run the cleaning process
cf.run()

# Print cleaning report
cf.report()

# Visualize clusters and frame pairs
cf.visualize_clusters()

Caching and Outputs

• Embeddings and cleaning reports are cached within the specified cache folder for faster reruns.
• Cleaned and removed images are saved beside the original frames in the dataset folder, allowing easy inspection and further use.
• The caching mechanism avoids redundant computations, improving efficiency when processing large datasets.

Supported Embedding Models

CleanFrames supports multiple embedding models for frame representation, including but not limited to:

• CLIP models such as clip-ViT-B-32 and Swin
• Additional models can be integrated as needed.

Clustering Methods

Available clustering algorithms include:

• KMeans clustering
• DBSCAN clustering
• Other clustering methods can be added or customized.

Visualization

CleanFrames provides visualization tools to help users inspect the clustering results and pairs of similar frames. This helps verify the cleaning quality and understand the grouping of frames.

Reporting

After cleaning, CleanFrames generates a concise text-only console report summarizing:

• Number of frames processed
• Number of frames removed
• Number of frames retained

This report provides insights into the effectiveness of the cleaning process.

---

For more detailed information and advanced usage, please refer to the source code and examples provided in the repository.