Omni-Modal Dissonance Benchmark: Systematically Breaking Modality Consensus to Probe Robustness and Calibrated Abstention
Provides a diagnostic benchmark for evaluating modality reliance, robustness, and uncertainty calibration in omni-modal systems.
Existing omni-modal benchmarks confound modality contributions due to correlated information. OMD-Bench introduces a benchmark with initially congruent modalities that are systematically corrupted to isolate each modality's contribution, finding that models over-abstain with two corrupted modalities but under-abstain with three, maintaining high confidence (60-100%) even under full corruption.
Existing omni-modal benchmarks attempt to measure modality-specific contributions, but their measurements are confounded: naturally co-occurring modalities carry correlated yet unequal information, making it unclear whether results reflect true modality reliance or information asymmetry. We introduce OMD-Bench, where all modalities are initially congruent - each presenting the same anchor, an object or event independently perceivable through video, audio, and text - which we then systematically corrupt to isolate each modality's contribution. We also evaluate calibrated abstention: whether models appropriately refrain from answering when evidence is conflicting. The benchmark comprises 4,080 instances spanning 27 anchors across eight corruption conditions. Evaluating ten omni-modal models under zero-shot and chain-of-thought prompting, we find that models over-abstain when two modalities are corrupted yet under-abstain severely when all three are, while maintaining high confidence (~60-100%) even under full corruption. Chain-of-thought prompting improves abstention alignment with human judgment but amplifies overconfidence rather than mitigating it. OMD-Bench provides a diagnostic benchmark for diagnosing modality reliance, robustness to cross-modal inconsistency, and uncertainty calibration in omni-modal systems.