D-CoDe: Scaling Image-Pretrained VLMs to Video via Dynamic Compression and Question Decomposition

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Huang, Yiyang, Wang, Yizhou, Fu, Yun
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866911201697464320
author Huang, Yiyang
Wang, Yizhou
Fu, Yun
author_facet Huang, Yiyang
Wang, Yizhou
Fu, Yun
contents Video large language models (Vid-LLMs), which excel in diverse video-language tasks, can be effectively constructed by adapting image-pretrained vision-language models (VLMs). However, this adaptation remains challenging, as it requires processing dense and temporally extended visual inputs that exceed the capacity of image-based models. This paper identifies the perception bottleneck and token overload as key challenges in extending image-based VLMs to the video domain. To address these issues, we propose D-CoDe, a training-free adaptation framework that incorporates dynamic compression and question decomposition. Specifically, dynamic compression alleviates the perception bottleneck through adaptive selection of representative frames and content-aware aggregation of spatial tokens, thereby reducing redundancy while preserving informative content. In parallel, question decomposition mitigates token overload by reformulating the original query into sub-questions, guiding the model to focus on distinct aspects of the video and enabling more comprehensive understanding. Experiments demonstrate that D-CoDe effectively improves video understanding across various benchmarks. Furthermore, strong performance on the challenging long-video benchmark highlights the potential of D-CoDe in handling complex video-language tasks. Code is available at https://github.com/hukcc/D-CoDe.
format Preprint
id arxiv_https___arxiv_org_abs_2510_08818
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle D-CoDe: Scaling Image-Pretrained VLMs to Video via Dynamic Compression and Question Decomposition
Huang, Yiyang
Wang, Yizhou
Fu, Yun
Computer Vision and Pattern Recognition
Artificial Intelligence
Video large language models (Vid-LLMs), which excel in diverse video-language tasks, can be effectively constructed by adapting image-pretrained vision-language models (VLMs). However, this adaptation remains challenging, as it requires processing dense and temporally extended visual inputs that exceed the capacity of image-based models. This paper identifies the perception bottleneck and token overload as key challenges in extending image-based VLMs to the video domain. To address these issues, we propose D-CoDe, a training-free adaptation framework that incorporates dynamic compression and question decomposition. Specifically, dynamic compression alleviates the perception bottleneck through adaptive selection of representative frames and content-aware aggregation of spatial tokens, thereby reducing redundancy while preserving informative content. In parallel, question decomposition mitigates token overload by reformulating the original query into sub-questions, guiding the model to focus on distinct aspects of the video and enabling more comprehensive understanding. Experiments demonstrate that D-CoDe effectively improves video understanding across various benchmarks. Furthermore, strong performance on the challenging long-video benchmark highlights the potential of D-CoDe in handling complex video-language tasks. Code is available at https://github.com/hukcc/D-CoDe.
title D-CoDe: Scaling Image-Pretrained VLMs to Video via Dynamic Compression and Question Decomposition
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2510.08818