dataset.md

March 6, 2025 · View on GitHub

Prepare EmbodiedScan Data

Given the licenses of respective raw datasets, we recommend users download the raw data from their official websites and then organize them following the below guide. Detailed steps are shown as follows.

Download ScanNet v2 data HERE. Link or move the folder to this level of directory.
Download ScanQA data HERE. Link or move the folder to this level of directory.
Download SQA data HERE. Link or move the folder to this level of directory. The directory structure should be as below.

data
├── scannet
│   ├── meta_data
│   ├── scans
│   │   ├── <scene_id>
│   │   ├── ...
│   ├── scans_test
│   │   ├── <scene_id>
│   │   ├── ...
├── qa
│   ├── ScanQA_v1.0_test_w_obj.json
│   ├── ...
├── sqa_task
│   ├── answer_dict.json
│   ├── balanced
│   │   ├── v1_balanced_questions_test_scannetv2.json
│   │   ├── ...

Preprocess scannet's point clounds:

cd ./data/scannet
python batch_load_scannet_data.py

Enter the project root directory, extract images:

python embodiedqa/converter/generate_image_scannetv2.py --dataset_folder ./data/scannet/ --fast

Preprocess scannet's annotation:

python embodiedqa/converter/create_scannetv2_info_pkl.py --dataset_folder ./data/scannet/ --output_dir ./data

The directory structure should be as below after that

data
├── scannet
│   ├── meta_data
│   ├── scans
│   │   ├── <scene_id>
│   │   ├── ...
│   ├── scans_test
│   │   ├── <scene_id>
│   │   ├── ...
│   ├── posed_images
│   │   ├── <scene_id>
│   │   |   ├── *.jpg
│   │   |   ├── *.png
│   │   ├── ...
│   ├── posed_images_test
│   │   ├── <scene_id>
│   │   |   ├── *.jpg
│   │   |   ├── *.png
│   │   ├── ...
│   ├── scannet_data
│   │   ├── <scene_id>_aligned_vert.npy
│   │   ├── ...
├── qa
│   ├── ScanQA_v1.0_test_w_obj.json
│   ├── ...
├── sqa_task
│   ├── answer_dict.json
│   ├── balanced
│   │   ├── v1_balanced_questions_test_scannetv2.json
│   │   ├── ...
├── mv_scannetv2_infos_test.pkl
├── mv_scannetv2_infos_train.pkl
├── mv_scannetv2_infos_val.pkl