Ego-Grounding for Personalized Question-Answering in Egocentric Videos

Dimensions overall score 7.0

• The MyEgo dataset is the first egocentric VideoQA dataset designed to evaluate multimodal large language models' ability to understand, remember, and reason about the camera wearer.method 95%
  Evidencepartial

  We introduce MyEgo, the first egocentric VideoQA dataset designed to evaluate MLLMs' ability to understand, remember, and reason about the camera wearer.

  Implicationpartial

  Explicitly stated in the abstract as 'the first egocentric VideoQA dataset designed to evaluate MLLMs' ability to understand, remember, and reason about the camera wearer'

  Verificationpartial

  partial

• Top closed-source models (e.g., GPT-5) achieve only about 46% accuracy on the MyEgo dataset, trailing human performance by nearly 40%.result 90%
  Evidencepartial

  Top closed- and open-source models (e.g., GPT-5 and Qwen3-VL) achieve only~46% and 36% accuracy, trailing human performance by near 40% and 50% respectively.

  Implicationpartial

  Direct numeric evidence provided in abstract with specific model names and performance metrics

  Verificationpartial

  partial

• Top open-source models (e.g., Qwen3-VL) achieve only about 36% accuracy on the MyEgo dataset, trailing human performance by nearly 50%.result 90%
  Evidencepartial

  Top closed- and open-source models (e.g., GPT-5 and Qwen3-VL) achieve only~46% and 36% accuracy, trailing human performance by near 40% and 50% respectively.

  Implicationpartial

  Direct numeric evidence provided in abstract with specific model names and performance metrics

  Verificationpartial

  partial

• Neither explicit reasoning nor model scaling yield consistent improvements in performance on the MyEgo dataset.result 85%
  Evidencepartial

  Surprisingly, neither explicit reasoning nor model scaling yield consistent improvements.

  Implicationpartial

  Directly stated in abstract as a finding from benchmarking, though specific evidence details would be in full paper

  Verificationpartial

  partial

• Models improve when relevant evidence is explicitly provided, but gains drop over time, indicating limitations in tracking and remembering 'me' and 'my past'.limitation 85%
  Evidencepartial

  Models improve when relevant evidence is explicitly provided, but gains drop over time, indicating limitations in tracking and remembering 'me' and 'my past'.

  Implicationpartial

  Directly stated in abstract as a key finding from the analysis

  Verificationpartial

  partial

• Competitive multimodal large language models across variants all struggle on the MyEgo dataset.result 80%
  Evidencepartial

  Benchmarking reveals that competitive MLLMs across variants, including open-source vs. proprietary, thinking vs. non-thinking, small vs. large scales all struggle on MyEgo.

  Implicationpartial

  Strongly supported by benchmarking results showing low accuracy across model types, though specific comparison details would be in full paper

  Verificationpartial

  partial

• Ego-grounding and long-range memory play crucial roles in enabling personalized question-answering in egocentric videos.technical 80%
  Evidencepartial

  These findings collectively highlight the crucial role of ego-grounding and long-range memory in enabling personalized QA in egocentric videos.

  Implicationpartial

  Directly stated as a conclusion from the findings, though supported by the evidence presented

  Verificationpartial

  partial

• The MyEgo dataset comprises 541 long videos and 5K personalized questions asking about 'my things', 'my activities', and 'my past'.method 95%
  Evidencepartial

  MyEgo comprises 541 long videos and 5K personalized questions asking about 'my things', 'my activities', and 'my past'.

  Implicationpartial

  Explicit numeric details provided about dataset composition and question types

  Verificationpartial

  partial

Competitive landscape

A new dataset and benchmark for personalized question-answering in egocentric videos, revealing significant limitations in current multimodal LLMs for understanding the camera wearer.