StreamGaze + EgoGazeVQA — combined gaze-grounded video QA datasetsA single repository containing two complementary benchmarks for evaluating
multimodal LLMs on gaze-grounded egocentric video question answering:
Subfolder
Source
Questions
Format
StreamGaze_v2/
egoexolearn, holoassist, egtea
8 MCQ tasks (4-opt) — gaze-conditioned past/present/future
streaming QA
EgoGazeVQA/
ego4d, egoexo, egtea
causal / spatial / temporal (5-opt)
per-clip QA
For training that uses… See the full description on the dataset page:
https://huggingface.co/datasets/Peanuttoad/StreamGaze_EgoGazeVQA.
