Do You See What I Am Pointing At? Gesture-Based Egocentric Video Question Answering

Dimensions overall score 8.0

GitHub Code Pulse

No public code linked for this paper yet.

• Current Multimodal Large Language Models (MLLMs) struggle with gesture-based egocentric question answering tasks due to lack of gesture-rich data and limited ability to infer fine-grained pointing intent.limitation 90%
  Evidencepartial

  current Multimodal Large Language Models (MLLMs) struggle with such tasks due to the lack of gesture-rich data and their limited ability to infer fine-grained pointing intent from egocentric video

  Implicationpartial

  Directly stated in the abstract as the motivation for the research

  Verificationpartial

  partial

• The EgoPointVQA dataset comprises 4000 synthetic and 400 real-world videos across multiple deictic reasoning tasks.method 95%
  Evidencepartial

  we introduce EgoPointVQA, a dataset and benchmark for gesture-grounded egocentric question answering, comprising 4000 synthetic and 400 real-world videos across multiple deictic reasoning tasks

  Implicationpartial

  Explicitly stated with specific numeric values in the abstract

  Verificationpartial

  partial

• Hand Intent Tokens (HINT) encode tokens derived from 3D hand keypoints using an off-the-shelf reconstruction model and interleave them with model input to provide explicit spatial and temporal context.method 90%
  Evidencepartial

  we further propose Hand Intent Tokens (HINT), which encodes tokens derived from 3D hand keypoints using an off-the-shelf reconstruction model and interleaves them with the model input to provide explicit spatial and temporal context for interpreting pointing intent

  Implicationpartial

  Directly described in the abstract with specific technical details

  Verificationpartial

  partial

• HINT-14B achieves 68.1% accuracy on average over 6 tasks, surpassing the state-of-the-art InternVL3-14B by 6.6%.result 95%
  Evidencepartial

  HINT-14B achieves 68.1% accuracy, on average over 6 tasks, surpassing the state-of-the-art, InternVL3-14B, by 6.6%

  Implicationpartial

  Explicitly stated with precise numeric results and comparison

  Verificationpartial

  partial

• The proposed model outperforms others in different backbones and model sizes.result 85%
  Evidencepartial

  We show that our model outperforms others in different backbones and model sizes

  Implicationpartial

  Directly stated in abstract but without specific comparison details

  Verificationpartial

  partial

• The code, model, and dataset will be released to facilitate open research.method 90%
  Evidencepartial

  To further facilitate the open research, we will release the code, model, and dataset

  Implicationpartial

  Explicit commitment stated in the abstract

  Verificationpartial

  partial

• Understanding and answering questions based on a user's pointing gesture is essential for next-generation egocentric AI assistants.market 80%
  Evidencepartial

  Understanding and answering questions based on a user's pointing gesture is essential for next-generation egocentric AI assistants

  Implicationpartial

  Presented as a foundational assumption but not empirically proven in the provided text

  Verificationpartial

  partial

• The approach provides explicit spatial and temporal context for interpreting pointing intent.technical 85%
  Evidencepartial

  interleaves them with the model input to provide explicit spatial and temporal context for interpreting pointing intent

  Implicationpartial

  Directly stated as a feature of the HINT method

  Verificationpartial

  partial

Startup potential card