How to Download the ADT Dataset
Overview
This page covers how to download the sample Aria Digital Twin (ADT) sequence as well as how to download specific sequences and types of data. Follow the instructions to download the sample datasets and from there you'll be able to use the CLI to download more data. For information on how to download ADT object models, see Object Models page
The sample dataset is a single-user dataset with body pose in the Apartment. This is a pretty representative example that should give you an idea of the dataset.
By downloading the datasets you agree that you have read and accepted the terms of the Aria Digital Twin Dataset License Agreement.
Download the sample Aria Digital Twin (ADT) sequence
Step 0: install project_aria_tools package and create venv if not done before
Follow Step 0 to Step 3 in Getting Started.
Step 1 : Signup and get the download links file
To use the downloader CLI, you need to download a file which contains all data URLs plus some metadata for the downloader script. We currently offer two ways of getting this file:
Option 1 - Aria Dataset Explorer:
Go to the Aria Dataset Explorer website. Here you can subselect sequences according to some filter options, or get the links to all sequences.
Option 2 - ADT Website:
Visit ADT website and sign up.
Scroll down to the bottom of the page. Enter your email and select Access the Datasets.
Once you've selected Access the Datasets you'll be taken back to the top of the ADT page.
Scroll down the page to select ADT Sequences link and download the file to the folder $HOME/Downloads.
You can redownload the download links whenever they expire
Step 2 : Set up a folder for ADT data
mkdir -p $HOME/Documents/projectaria_tools_adt_data
Step 3 : Download the sample sequence (~500MB) via CLI:
From your Python virtual environment, run:
aria_dataset_downloader -c ${PATH_TO_YOUR_CDN_FILE} \
-o $HOME/Documents/projectaria_tools_adt_data/ \
-l Apartment_release_golden_skeleton_seq100_10s_sample_M1292 \
-d 0 1 2 3 4 5 6 7 8 9
The sample dataset is a single-user dataset with body pose in the Apartment. This is a pretty representative example to give an idea of the dataset. For more information on the content in the other sequences, see the Data Content section below
Download the Aria Digital Twin (ADT) benchmark dataset
Data size
The Aria Digital Twin dataset consists of 236 sequences. It is split into 4 data types that can be downloaded individually, plus MPS data. Go to Project Aria Machine Perception Services for more information about MPS data. The MPS data is also broken into chunks that can be included or excluded at download time. The size of each data type is shown below. ADT data without MPS is approximately 3.5TB.
| Data type | What's included | Per sequence size | Total size for all sequences |
|---|---|---|---|
| main | Aria raw data, 2D bounding box, 3D object poses and bounding box, skeleton data, eye gaze data | 3 - 6 GB | ~700 GB |
| segmentation | Instance segmentation data | 2 - 4 GB | ~750 GB |
| depth | Depth map data | 4 - 8 GB | ~1.5 TB |
| synthetic | Synthetic rendering data | 2 - 4 GB | 500 GB |
| MPS eyegaze | Eyegaze, summary file | < 1 MB | ~31 MB |
| MPS SLAM points | Semi-dense points and observations | 200 - 500 MB | ~31 GB |
| MPS SLAM trajectories | Open and closed loop trajectories | 100 - 200 MB | 12 GB |
| MPS SLAM online calibration | Online calibrations | < 20 MB | 1.2 GB |
Download via CLI
ADT uses the general Aria dataset downloader, which is available in the projectaria_tools PyPI (pip install) package.
To use the downloader, use the following commands in the virtual environment where you've installed projectaria_tools:
aria_dataset_downloader
To get a description of the arguments that the script needs, run:
aria_dataset_downloader --help
The following are some example use cases:
Download VRS for all sequences
aria_dataset_downloader --cdn_file ${PATH_TO_YOUR_CDN_FILE} --output_folder ${OUTPUT_FOLDER_PATH} --data_types 0
Download VRS + main ground truth data for all sequences
aria_dataset_downloader --cdn_file ${PATH_TO_YOUR_CDN_FILE} --output_folder ${OUTPUT_FOLDER_PATH} --data_types 0 6
Download all data for all sequences
aria_dataset_downloader --cdn_file ${PATH_TO_YOUR_CDN_FILE} --output_folder ${OUTPUT_FOLDER_PATH} --data_types 0 1 2 3 4 5 6 7 8 9
Download VRS for 2 specific sequences
aria_dataset_downloader --cdn_file ${PATH_TO_YOUR_CDN_FILE} --output_folder ${OUTPUT_FOLDER_PATH} --data_types 0 --sequence_names Apartment_release_clean_seq131_M1292 Apartment_release_clean_seq133_M1292
Download VRS for all sequences and overwrite
aria_dataset_downloader --cdn_file ${PATH_TO_YOUR_CDN_FILE} --output_folder ${OUTPUT_FOLDER_PATH} --data_types 0 --overwrite
Deprecation Notes
Dataset Metadata
We previously provided a dataset metadata JSON “aria_digital_twin_benchmark_metadata.json”, which was downloadable with the old adt_benchmark_dataset_downloader. The goal of this file was to allow users to filter sequences by data fields provided in each sequence metadata. The Aria Dataset Explorer has replaced this old method for filtering sequences.
Challenge Data
We will be removing access to the ADT challenge data now that the challenge has concluded. The challenge downloader will be available until the end of 2024. You can find instructions for using this downloader on the ADT Challenge Website. To use this downloader, however, you will need the deprecated CDN links file. Please email us at ariaoperations@meta.com before December 31 2024 to get access to this metadata file.
Troubleshooting
Check Project Aria Tools troubleshooting if you are having issues using this guide.