MVTamperBench: Evaluating Robustness of Vision-Language Models

TL;DR

MVTamperBench is a comprehensive benchmark designed to evaluate the robustness of Multimodal Large Language Models against common visual tampering techniques, highlighting variability in resilience and emphasizing the need for tamper-resistant models.

Contribution

The paper introduces MVTamperBench, a new benchmark with extensive tampering scenarios to systematically assess MLLM robustness, addressing a critical gap in trustworthy video understanding.

Findings

Significant variability in model resilience across tampering types.

Larger models do not always exhibit greater robustness.

Benchmark facilitates development of tamper-resilient MLLMs for safety-critical applications.

Abstract

Multimodal Large Language Models (MLLMs), are recent advancement of Vision-Language Models (VLMs) that have driven major advances in video understanding. However, their vulnerability to adversarial tampering and manipulations remains underexplored. To address this gap, we introduce \textbf{MVTamperBench}, a benchmark that systematically evaluates MLLM robustness against five prevalent tampering techniques: rotation, masking, substitution, repetition, and dropping; based on real-world visual tampering scenarios such as surveillance interference, social media content edits, and misinformation injection. MVTamperBench comprises ~3.4K original videos, expanded into over ~17K tampered clips covering 19 distinct video manipulation tasks. This benchmark challenges models to detect manipulations in spatial and temporal coherence. We evaluate 45 recent MLLMs from 15+ model families. We reveal substantial variability in resilience across tampering types and show that larger parameter counts do not necessarily guarantee robustness. MVTamperBench sets a new benchmark for developing tamper-resilient MLLM in safety-critical applications, including detecting clickbait, preventing harmful content distribution, and enforcing policies on media platforms. We release all code, data, and benchmark to foster open research in trustworthy video understanding. Code: https://amitbcp.github.io/MVTamperBench/ Data: https://huggingface.co/datasets/Srikant86/MVTamperBench