Video binary classification using deep learning techniques

Summary

In the video summarization domain it is needed to efficiently differentiate between informative and non-informative video segments to create concise summaries that encapsulate essential content. Utilizing advanced deep learning methods for feature extraction from both audio and visual data, the study employs a diverse array of optimized classification algorithms and novel LSTM, alongside Attention-based models and Transformers. An early fusion approach integrates audio-visual data to enhance classification accuracy. Despite notable successes, particularly with visual data, challenges in audio feature extraction and certain model performances indicate areas for future improvement. The thesis contributes to the field by demonstrating the potential of combining aural and visual features using deep learning techniques for video binary classification, setting a solid groundwork for advancements in achieving more accurate video summarizations.