PAVE: Patching and Adapting Video Large Language Models
This work addresses the problem of efficiently extending video large language models to handle diverse multimodal tasks for researchers and practitioners, representing an incremental improvement through adapter-based adaptation.
The paper tackles the challenge of adapting pre-trained video large language models to new tasks involving additional modalities like audio or 3D information, and presents PAVE, a framework that introduces lightweight adapters to enhance performance across tasks such as audio-visual question answering and 3D reasoning, achieving state-of-the-art results with only ~0.1% additional computational cost.
Pre-trained video large language models (Video LLMs) exhibit remarkable reasoning capabilities, yet adapting these models to new tasks involving additional modalities or data types (e.g., audio or 3D information) remains challenging. In this paper, we present PAVE, a flexible framework for adapting pre-trained Video LLMs to downstream tasks with side-channel signals, such as audio, 3D cues, or multi-view videos. PAVE introduces lightweight adapters, referred to as "patches," which add a small number of parameters and operations to a base model without modifying its architecture or pre-trained weights. In doing so, PAVE can effectively adapt the pre-trained base model to support diverse downstream tasks, including audio-visual question answering, 3D reasoning, multi-view video recognition, and high frame rate video understanding. Across these tasks, PAVE significantly enhances the performance of the base model, surpassing state-of-the-art task-specific models while incurring a minor cost of ~0.1% additional FLOPs and parameters. Further, PAVE supports multi-task learning and generalizes well across different Video LLMs. Our code is available at https://github.com/dragonlzm/PAVE.