Please use this identifier to cite or link to this item: http://hdl.handle.net/10609/149808
Title: VClipper: Exploiting CLIP Zero-shot capabilities for moment retrieval in video recordings
Author: Caravaca Muller, Oriol
Tutor: Benito Altamirano, Ismael  
Abstract: This research explores the integration of CLIP, a pretrained model, into video content analysis. In a landscape inundated with multimedia data, pinpointing specific moments within videos is a persistent challenge. By leveraging CLIP's semantic and visual search capabilities, this study endeavors to refine content retrieval methods. Emphasizing efficiency and applicability, this study aims to make this process more precise and practical. With this research we also reviewed the state-of-the-art methods and produced empirical analysis on the effects of postprocessing on the similarity vectors obtained from CLIP encoders. Finally, we developed two distinct methods aimed at moment retrieval tasks in audiovisual data, obtaining a model that is able to outperform previous works in Zero-shot moment revival, reaching 57.3 at R@1 IoU=0.5 and 51.6 at mAP@0.5.
Keywords: video analysis
moment retrieval
CLIP
Document type: info:eu-repo/semantics/masterThesis
Issue Date: 9-Jan-2024
Publication license: http://creativecommons.org/licenses/by-nc-nd/3.0/es/  
Appears in Collections:Bachelor thesis, research projects, etc.

Files in This Item:
File Description SizeFormat 
ocaravacamTFM0123memorioa.pdfReport of FMDP2,19 MBAdobe PDFThumbnail
View/Open
Share:
Export:
View statistics

This item is licensed under aCreative Commons License Creative Commons