This toolkit automates downloading, analyzing, tagging, and classifying videos. It leverages the Qwen2-VL multimodal AI model for content understanding, along with traditional video processing techniques, to provide rich metadata and insights.

• Video Download: Downloads videos from various sources (direct links and HTML pages).
• Metadata Extraction: Extracts technical metadata using FFmpeg (duration, resolution, codecs, etc.).
• Scene Detection: Identifies scene changes for granular video analysis.
• AI-Powered Analysis: Analyzes video content using the Qwen2-VL model, generating descriptions and insights.
• Customizable Tagging: Applies tags based on metadata, Qwen2-VL output, and custom rules.
• Video Classification: Classifies videos into categories (e.g., sports, music).
• Data Export: Exports processed data in JSON format.
• MongoDB Integration (Optional): Stores data in a MongoDB database.

• Python 3.7 or higher: https://www.python.org/downloads/
• FFmpeg: Install FFmpeg on your system. Instructions vary by operating system.
• MongoDB (Optional): If you want to use MongoDB for storage, install it and set up a database.

1. Clone the repository:

   git clone https://github.com/henryperkins/video_processing_toolkit.git
   cd video_processing_toolkit

2. Create a virtual environment (recommended):

   python3 -m venv .venv
   source .venv/bin/activate

3. Install dependencies:

   pip install -r requirements.txt

4. Download the Qwen2-VL Model:

   • Download the model files from Hugging Face: https://huggingface.co/Qwen/Qwen2-VL-7B-Instruct
   • Extract the model files to the directory specified in your config.ini file (see below).

5. Configure config.ini:

   • Create a config.ini file in the project directory and customize the settings:

     [Paths]
     DownloadDirectory = downloaded_videos
     ProcessedDirectory = processed_videos
     
     [MongoDB]
     URI = your_mongodb_uri  ; Replace with your actual MongoDB URI
     DatabaseName = your_database_name
     CollectionName = your_collection_name
     
     [Qwen]
     ModelDir = /path/to/your/qwen2-vl-model  ; Replace with the path to the extracted model files
     
     [Concurrency]
     MaxWorkers = 5
     
     [SceneDetection]
     DefaultThreshold = 30.0
     
     [QwenVL]
     DefaultInstruction = Describe this video.

1. Run the script:

   python interface.py --urls <video_url_1> <video_url_2> ... --output <output_directory>

   • Replace <video_url_1>, <video_url_2>, etc. with the URLs of the videos you want to process.
   • Replace <output_directory> with the directory where you want to save the processed JSON files.

2. (Optional) Use a CSV file:

   • Create a CSV file with columns file_name, file_size, last_modified, and public_url containing video information.
   • Run the script with the --csv option:

     python interface.py --csv <path_to_csv_file> --output <output_directory>

The toolkit generates a JSON file for each processed video in the specified output directory. The JSON file contains:

• Video metadata (duration, resolution, codec, etc.)
• Scene change timestamps
• Qwen2-VL generated description
• Applied tags
• Video classification

• Process videos from URLs:

  python interface.py --urls https://www.example.com/video1.mp4 https://www.example.com/video2.mp4 --output processed_videos

• Process videos from a CSV file:

  python interface.py --csv video_list.csv --output processed_videos

Contributions are welcome! Please feel free to submit a pull request.

This project is licensed under the MIT License.