AVTrustBench: Assessing and Enhancing Reliability and Robustness in Audio-Visual LLMs

TL;DR

AVTrustBench is a comprehensive benchmark with 600K samples designed to evaluate and improve the reliability and robustness of audio-visual large language models across adversarial, compositional, and modality-specific tasks.

Contribution

The paper introduces AVTrustBench, the first extensive AV multi-task benchmark, and proposes CAVPref, a calibration training strategy to enhance AVLLMs' robustness and reliability.

Findings

Most existing AVLLMs perform poorly on human-like comprehension.

CAVPref improves model robustness by up to 30.19%.

Benchmark and code will be publicly released.

Abstract

With the rapid advancement of Multi-modal Large Language Models (MLLMs), several diagnostic benchmarks have recently been developed to assess these models' multi-modal reasoning proficiency. However, these benchmarks are restricted to assessing primarily the visual aspect and do not examine the holistic audio-visual (AV) understanding. Moreover, currently, there are no benchmarks that investigate the capabilities of AVLLMs to calibrate their responses when presented with perturbed inputs. To this end, we introduce Audio-Visual Trustworthiness assessment Benchmark (AVTrustBench), comprising 600K samples spanning over 9 meticulously crafted tasks, evaluating the capabilities of AVLLMs across three distinct dimensions: Adversarial attack, Compositional reasoning, and Modality-specific dependency. Using our benchmark we extensively evaluate 13 state-of-the-art AVLLMs. The findings reveal that the majority of existing models fall significantly short of achieving human-like comprehension, offering valuable insights for future research directions. To alleviate the limitations in the existing approaches, we further propose a robust, model-agnostic calibrated audio-visual preference optimization based training strategy CAVPref, obtaining a gain up to 30.19% across all 9 tasks. We will publicly release our code and benchmark to facilitate future research in this direction.