Breaking the Encoder Barrier for Seamless Video-Language Understanding

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Li, Handong, Zhang, Yiyuan, Guo, Longteng, Yue, Xiangyu, Liu, Jing
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909887221465088
author Li, Handong
Zhang, Yiyuan
Guo, Longteng
Yue, Xiangyu
Liu, Jing
author_facet Li, Handong
Zhang, Yiyuan
Guo, Longteng
Yue, Xiangyu
Liu, Jing
contents Most Video-Large Language Models (Video-LLMs) adopt an encoder-decoder framework, where a vision encoder extracts frame-wise features for processing by a language model. However, this approach incurs high computational costs, introduces resolution biases, and struggles to capture fine-grained multimodal interactions. To overcome these limitations, we propose ELVA, an encoder-free Video-LLM that directly models nuanced video-language interactions without relying on a vision encoder. ELVA employs token merging to construct a bottom-up hierarchical representation and incorporates a video guidance supervisor for direct spatiotemporal representation learning. Additionally, a hybrid-resolution mechanism strategically integrates high- and low-resolution frames as inputs to achieve an optimal balance between performance and efficiency. With only 7M publicly available video-text pairs, ELVA achieves performance on par with encoder-based Video-LLMs while reducing FLOPs by up to 95\% and inference latency by 92\%, offering a scalable and efficient solution for real-time video understanding.
format Preprint
id arxiv_https___arxiv_org_abs_2503_18422
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Breaking the Encoder Barrier for Seamless Video-Language Understanding
Li, Handong
Zhang, Yiyuan
Guo, Longteng
Yue, Xiangyu
Liu, Jing
Computer Vision and Pattern Recognition
Most Video-Large Language Models (Video-LLMs) adopt an encoder-decoder framework, where a vision encoder extracts frame-wise features for processing by a language model. However, this approach incurs high computational costs, introduces resolution biases, and struggles to capture fine-grained multimodal interactions. To overcome these limitations, we propose ELVA, an encoder-free Video-LLM that directly models nuanced video-language interactions without relying on a vision encoder. ELVA employs token merging to construct a bottom-up hierarchical representation and incorporates a video guidance supervisor for direct spatiotemporal representation learning. Additionally, a hybrid-resolution mechanism strategically integrates high- and low-resolution frames as inputs to achieve an optimal balance between performance and efficiency. With only 7M publicly available video-text pairs, ELVA achieves performance on par with encoder-based Video-LLMs while reducing FLOPs by up to 95\% and inference latency by 92\%, offering a scalable and efficient solution for real-time video understanding.
title Breaking the Encoder Barrier for Seamless Video-Language Understanding
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2503.18422