Zero-Shot and Few-Shot Video Question Answering with Multi-Modal Prompts

TL;DR

This paper proposes a parameter-efficient multimodal prompt learning method for zero-shot and few-shot video question answering, effectively leveraging large-scale pretrained models while addressing overfitting and cross-modal gaps.

Contribution

It introduces a novel approach combining multimodal prompts and a transformer-based mapping network, maintaining frozen pretrained models for improved performance and efficiency.

Findings

Outperforms existing methods on multiple video QA benchmarks

Effective in zero-shot and few-shot scenarios

Reduces parameter count while maintaining high accuracy

Abstract

Recent vision-language models are driven by large-scale pretrained models. However, adapting pretrained models on limited data presents challenges such as overfitting, catastrophic forgetting, and the cross-modal gap between vision and language. We introduce a parameter-efficient method to address these challenges, combining multimodal prompt learning and a transformer-based mapping network, while keeping the pretrained models frozen. Our experiments on several video question answering benchmarks demonstrate the superiority of our approach in terms of performance and parameter efficiency on both zero-shot and few-shot settings. Our code is available at https://engindeniz.github.io/vitis.