Enhancing Text-to-Video Retrieval Using Clip Based Deep Learning Approach

Text-to-video retrieval is an essential task in multimedia information retrieval, enabling users to search and retrieve videos based on natural language descriptions. In this paper, we explore a deep learning approach leveraging the CLIP model for feature extraction and retrieval on the MSR-VTT dataset. We sample video frames and extract joint visual and textual embeddings using attention-based fusion and L2 normalization. Our basic model achieves Recall@1 of 13.96%, Recall@5 of 28.63%, Recall@10 of 36.40%, and Recall@50 of 57.12%, with a Median Rank (MdR) of 30 and a Mean Rank (MnR) of 406.75. We further discuss limitations and propose improvements such as multi-modal fusion, fine-tuning of the CLIP model, and advanced attention mechanisms. Our findings are compared against state-of-the-art methods, highlighting both progress and remaining challenges.