EgoTaskQA
We provide a basic illustration for all data here, in this directory we have
|-- metadata
|-- actions.json # Original LEMMA metadata
|-- hois.json # Original LEMMA metadata
|-- lengths.json # Video lengths in summary for original LEMMA videos
|-- objects.json # Original LEMMA metadata
|-- relationships.json # Augmented LEMMA relationships metadata
|-- states.json # Augmented LEMMA states metadata
|-- tasks.json # Original LEMMA metadata
|-- qa
|-- direct # Normal split of EgoTaskQA
|-- train_qas.json
|-- val_qas.json
|-- test_qas.json
|-- indirect # Indirect split of EgoTaskQA
|-- train_qas.json
|-- val_qas.json
|-- test_qas.json
|-- qa_videos.zip # Short clips cropped from LEMMA for EgoTaskQA
|-- state_annotations.json # World state annotations in augmented LEMMA dataset