ELV-Halluc: Benchmarking Semantic Aggregation Hallucinations in Long Video Understanding

2025-08-29

AI Summary

Video multimodal large language models (Video-MLLMs) suffer from Semantic Aggregation Hallucination (SAH) in long-video understanding—erroneously aggregating frame-level semantics into inconsistent event-level interpretations misaligned with ground-truth video content. Existing hallucination benchmarks focus on short videos and fail to characterize SAH’s root causes or its exacerbation with increasing semantic complexity and temporal dynamics. Method: We formally define and isolate SAH, introducing ELV-Halluc—the first benchmark dedicated to evaluating hallucinations in long-video understanding. We propose three complementary techniques: adversarial data construction, optimized positional encoding, and DPO-based alignment training to enhance model discrimination between intra- and inter-event semantics. Contribution/Results: Evaluated on 8K adversarial samples, our approach reduces SAH rate by 27.7% on both ELV-Halluc and Video-MME, significantly improving robustness and accuracy in long-video comprehension.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Large Vision Models
• Natural Language Processing: (Large) Language Models

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Large language models for search
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling

Abstract

Video multimodal large language models (Video-MLLMs) have achieved remarkable progress in video understanding. However, they remain vulnerable to hallucination-producing content inconsistent with or unrelated to video inputs. Previous video hallucination benchmarks primarily focus on short-videos. They attribute hallucinations to factors such as strong language priors, missing frames, or vision-language biases introduced by the visual encoder. While these causes indeed account for most hallucinations in short videos, they still oversimplify the cause of hallucinations. Sometimes, models generate incorrect outputs but with correct frame-level semantics. We refer to this type of hallucination as Semantic Aggregation Hallucination (SAH), which arises during the process of aggregating frame-level semantics into event-level semantic groups. Given that SAH becomes particularly critical in long videos due to increased semantic complexity across multiple events, it is essential to separate and thoroughly investigate the causes of this type of hallucination. To address the above issues, we introduce ELV-Halluc, the first benchmark dedicated to long-video hallucination, enabling a systematic investigation of SAH. Our experiments confirm the existence of SAH and show that it increases with semantic complexity. Additionally, we find that models are more prone to SAH on rapidly changing semantics. Moreover, we discuss potential approaches to mitigate SAH. We demonstrate that positional encoding strategy contributes to alleviating SAH, and further adopt DPO strategy to enhance the model's ability to distinguish semantics within and across events. To support this, we curate a dataset of 8K adversarial data pairs and achieve improvements on both ELV-Halluc and Video-MME, including a substantial 27.7% reduction in SAH ratio.

Problem

• Benchmarking semantic aggregation hallucinations in long video understanding
• Investigating incorrect outputs despite correct frame-level semantics
• Addressing hallucinations from aggregating frame to event semantics

Innovation

• Introduces ELV-Halluc benchmark for long-video hallucination analysis
• Uses positional encoding to reduce semantic aggregation errors
• Applies DPO strategy with 8K adversarial pairs for improvement