metadata
license: mit
task_categories:
- video-classification
- question-answering
tags:
- egocentric-video
- action-planning
- video-qa
size_categories:
- n<1K
EgoPlan-Bench Dataset
This dataset contains 741 samples from the EgoPlan-Bench benchmark for egocentric video understanding and action planning.
Dataset Structure
Each sample contains:
- video: The video file
- question: The question about what action should be taken next
- task_goal: The overall task goal
- options: List of 4 possible actions (choices a, b, c, d)
- answer: The correct answer (a, b, c, or d)
- sample_id: Original sample ID
- video_id: Original video ID
Usage
from datasets import load_dataset
dataset = load_dataset("advaitgupta/egoplan-bench")
# Access a sample
sample = dataset['train'][0]
print(sample['question'])
print(sample['options'])
print(sample['answer'])
Citation
If you use this dataset, please cite the original EgoPlan-Bench paper.
License
MIT License