Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
September 28, 2025Open Access

EGOILLUSION: Benchmarking Hallucinations in Egocentric Video Understanding

View Full Paper
Ask AI
Bookmark
Share

Authors

ASAshish SethIndian Institute of Technology MadrasUTUtkarsh TyagiIndian Space Research OrganisationRSR. K. SelvakumarSamsung (United States)

Discussion

Loading...

Member takes

Implication

EgoIllusion benchmarks hallucinations in MLLMs with 1,400 videos and 8,000 questions, indicating significant challenges in visual and auditory understanding.

Key Points

  • Only 59% accuracy was achieved by powerful models like GPT-4o and Gemini in hallucination evaluation.
  • EgoIllusion consists of 1,400 videos with 8,000 human-annotated questions to assess hallucinations.
  • This benchmarking approach aims to improve multimodal large language models in egocentric video contexts.
  • Open-sourcing the EgoIllusion dataset promotes reproducibility and encourages advancements in MLLMs.

Cite This Study

Seth et al. (2025) studied this question.

synapsesocial.com/papers/68d913a34ddcf71ba560bb0ehttps://doi.org/10.48550/arxiv.2508.12687
View Full Paper
Ask AI
Bookmark
Share

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

  1. 1VidHalluc: Evaluating Temporal Hallucinations in Multimodal Large Language Models for Video Understanding2025 · 13 citations
  2. 2VideoHallucer: Evaluating Intrinsic and Extrinsic Hallucinations in Large Video-Language Models2024 · 4 citations
  3. 3Hallucination of Multimodal Large Language Models: A Survey2024 · 30 citations
  4. 4GHOST: Hallucination-Inducing Image Generation for Multimodal LLMs2025
  5. 5Visual Hallucinations of Multi-modal Large Language Models2024