LLaVA-Octopus: Unlocking Instruction-Driven Adaptive Projector Fusion for Video Understanding

TL;DR

LLaVA-Octopus is a new video multimodal large language model that adaptively combines features from different visual projectors based on user instructions, improving performance in various video understanding tasks.

Contribution

It introduces a method to dynamically weight and fuse features from multiple visual projectors according to user instructions, enhancing multimodal video understanding.

Findings

Achieves state-of-the-art results on multiple video understanding benchmarks.

Effectively handles tasks like video question answering and long video comprehension.

Demonstrates broad applicability across diverse multimodal tasks.

Abstract

In this paper, we introduce LLaVA-Octopus, a novel video multimodal large language model. LLaVA-Octopus adaptively weights features from different visual projectors based on user instructions, enabling us to leverage the complementary strengths of each projector. We observe that different visual projectors exhibit distinct characteristics when handling specific tasks. For instance, some projectors excel at capturing static details, while others are more effective at processing temporal information, and some are better suited for tasks requiring temporal coherence. By dynamically adjusting feature weights according to user instructions, LLaVA-Octopus dynamically selects and combines the most suitable features, significantly enhancing the model's performance in multimodal tasks. Experimental results demonstrate that LLaVA-Octopus achieves excellent performance across multiple benchmarks, especially in tasks such as video question answering, long video understanding, and comprehensive multi-choices benchmarks, highlighting its broad application potential.