Due to the Habitat-Sim requirements, you must run this script on a Linux GPU-enabled machine.
The following requirements must be met to run the prepare_hm3d_dataset.py script, which generates the Ego4D VQA dataset:
- Before running this script, create a Conda environment following the instructions listed here.
- OpenEQA dataset, download it here, stored in the file open-eqa-v0.json.
- HM3D, which can be downloaded following the Habitat-Sim documentation.
Perform the following steps, executing from the hm3d directory:
python ./prepare_hm3d_dataset.py \
--openeqa_dataset [relative path to open-eqa-v0.json] \
--egoclip_metadata [relative path to the path of Matterport data] \
--video_output_dir [path of generated videos] \
--annotations_filename [path to output JSON]