Enrico Chen
Recursive Video Language Summarization Model: An Approach to Long Video Summarization.
Rel. Luca Cagliero, Lorenzo Vaiani, Alessia Intini. Politecnico di Torino, Corso di laurea magistrale in Data Science And Engineering, 2026
|
|
PDF (Tesi_di_laurea)
- Tesi
Accesso limitato a: Solo utenti staff fino al 24 Gennaio 2028 (data di embargo). Licenza: Creative Commons Attribution Non-commercial No Derivatives. Download (10MB) |
Abstract
Video summarization is a task in Natural Language Processing that consists in generating a concise summary of a video. In the case of Video-to-Text summarization, the summary takes the form of a textual description. Historically, extractive-based approaches have predominated, prioritizing the identification of salient frames and temporal segments of the input content. More recently, the focus has shifted toward abstractive approaches, which generate summaries that can include phrases not present in the original videos. Although Video Large Language Models (VideoLLMs) have demonstrated notable achievements in summarizing short videos, they continue to struggle with longer ones, producing suboptimal summaries due to challenges such as information preservation, context saturation limits, and temporal coherence degradation.
Conventional methods commonly rely on sparse frame sampling or fixed temporal segmentation, which inherently fragment event sequences and impede comprehensive reasoning over extended temporal ranges
Relatori
Anno Accademico
Tipo di pubblicazione
Numero di pagine
Corso di laurea
Classe di laurea
URI
![]() |
Modifica (riservato agli operatori) |
