Add video support for LIME in a new LimeVideoExplainer and VideoExplanation classes.

[sirTomasson]

Add Video Explanation Support to LIME

Overview

This PR introduces video explanation capabilities to LIME through a new LimeVideoExplainer class, enabling the interpretation of video classification models. This addition extends LIME's capabilities beyond images, text, and tabular data to address the growing need for explainability in video-based machine learning models.

Key Changes

• Added LimeVideoExplainer explainer class that implements LIME's methodology for video inputs
• Created example in a new Jupyter notebook documenting the usage of LimeVideoExplainer and the PyTorch framework
• Integrated with existing LIME architecture while maintaining code style consistency and without adding any new dependencies.

Why This Matters

Video classification models are increasingly important in applications like action recognition, surveillance, and content moderation. However, explaining their decisions has been challenging. This implementation helps:

• Researchers and practitioners understand why their video models make specific predictions
• End-users gain trust in video classification systems through transparent explanations
• Developers debug and improve video model performance by identifying relevant temporal and spatial features

Documentation

The included notebook demonstrates:

• Basic usage of LimeVideoExplainer
• Visualization of temporal and spatial importance in video segments
• Integration with popular video classification frameworks