Custom CNN-BiLSTM model for video captioning

This paper introduces a video captioning model that integrates spatial and temporal feature extraction methods to produce comprehensive textual descriptions for videos. The contributions of the study are detailed, starting with dataset preparation, where data collection, annotation, and preprocessing steps are explained to ensure a high-quality dataset for training and evaluation. The architecture of the video captioning model is then presented, incorporating Convolutional Neural Networks (CNN) for spatial feature extraction and Fast Fourier Transform (FFT) for temporal feature analysis. The extracted features are integrated into a Bidirectional Long Short-Term Memory (BiLSTM)-based text processing model to generate captions that effectively capture the associations between visual content and textual descriptions. The proposed approach has been proven effective through experimental results, showcasing enhanced BLEU scores and surpassing the performance of existing state-of-the-art models. The study concludes by suggesting future research directions and discussing potential advancements in video captioning. The contributions of this work pave the way for more advanced video captioning systems, enabling seamless integration between computer vision and natural language processing domains. Overall, this paper offers valuable insights into the dataset preparation process, the architecture of the video captioning model, experimental results, and future prospects for the proposed approach, contributing to the growing field of video captioning and its applications.