Download

Abstract

The preferred download method is via the Open Storage Network (OSN), which is a NSF-funded distributed data sharing and transfer service based on S3 buckets. We currently have a storage allocation on OSN which we are trialing to distribute I/Q-1M.

Get Access Keys

The I/Q-1M dataset is fully open-source and permissively licensed, and does not include any usage restrictions. To help us track usage statistics — and better target and motivate future datasets and extensions — we do ask that you fill out this short questionnaire to get the access keys for our dataset.

Set Up rclone

The best way to interact with OSN is through rclone.

Info

The I/Q-1m dataset is currently in pre-release; contact tianshu2@andrew.cmu.edu for the access keys.

After installing sudo apt-get install rclone, you will need to create a configuration with our bucket information:

• Find the config file name using rclone config file.

  Info

  The config path should be something like ~/.config/rclone/rclone.config i.e., /home/<user>/.config/rclone/rclone.config.

• Add the following to the file (create the file if it doesn't already exist):

  [osn-ro]
  type = s3
  provider = Ceph
  access_key_id =  # ...
  secret_access_key =   # ...
  endpoint = https://uri.osn.mghpcc.org
  no_check_bucket = true
  

Test the configuration

To test this configuration:

List filesDownload a sample file

$ rclone lsd osn-ro:/cmu-wiselab-iq1m
        0 2025-06-25 16:24:51        -1 data

lsd lists all directories. You can also use:

• rclone ls: list all files, recursively
• rclone ls --max-depth 1: list only files in the specified folder

rclone copy osn-ro:/cmu-wiselab-iq1m/data/bike/bloomfield.back/config.yaml .

Download from OSN

To download the full dataset (3.2TB / 2.9TiB), use:

rclone sync osn-ro:/cmu-wiselab-iq1m/data ./iq1m -v

You can also find our pre-trained model checkpoints in the same bucket:

rclone sync osn-ro:/cmu-wiselab-iq1m/checkpoints ./iq1m-checkpoints -v

Checkpoints

The following checkpoints are currently available:

$ rclone lsd osn-ro:/cmu-wiselab-iq1m/checkpoints
       0 2025-11-22 12:22:46        -1 base     # base model - 3D occupancy
       0 2025-11-22 12:22:46        -1 occ2d    # fine-tune - 2D occupancy
       0 2025-11-22 12:22:46        -1 semseg   # fine-tune - semantic segmentation
       0 2025-11-22 12:22:46        -1 vel      # fine-tune - ego-motion

Tip

When using rclone sync, you can stop (interrupt with ctrl+C) and resume downloading at any time.

Symlink Videos

The videos distributed via OSN have been anonymized to blur out faces; you can symlink these in place of the raw videos:

for i in `uv run roverd list /path/to/dataset`; do
    ln -s /path/to/dataset/$i/_camera/video.avi /path/to/dataset/$i/camera/video.avi;
done

Download Traces Individually

You can also download each setting or trace separately:

rclone sync osn-ro:/cmu-wiselab-iq1m/data/indoor ./iq1m/indoor -v
rclone sync osn-ro:/cmu-wiselab-iq1m/data/indoor/baker ./iq1m/indoor/baker -v
rclone sync osn-ro:/cmu-wiselab-iq1m/data/indoor/baker/baker.1.fwd ./iq1m/indoor/baker/baker.1.fwd -v

Exclude Unneeded Files

You can exclude certain files (e.g., lidar reflectance / NIR) using

rclone sync osn-ro:/cmu-wiselab-iq1m/data ./iq1m -v --exclude */lidar/rfl */lidar/nir

See the rclone sync documentation for more details and other options.

FTP Fallback

If you are unable to download the dataset via OSN, we have set up a "fallback" FTP server which also hosts a copy of the dataset.

Warning

This FTP server is hosted in our lab, and has limited resources compared to OSN; to avoid interfering with other traffic, the server is also rate-limited to 100Mbps. Please only use this option if OSN is down or otherwise inaccessible.

You can download files from the FTP server using any standard FTP client, e.g., FileZilla.

URL	Port
arena-gw.lan.cmu.edu	37285

Warning

The FTP server address and port are subject to change, and our lab network is explicitly not a high-availability service!

Verify Files

After downloading the dataset, you may wish to verify the integrity of the files by comparing checksums.

Info

You can find the reference checksum files in the red-rover repository. See roverd checksum for more details about how these checksums are computed and formatted; the format should be fairly self-explanatory once you untar the files.

Assuming that you've installed roverd into your environment:

1. Calculate checksums on your downloaded copy:

   for i in `roverd list /path/to/downloaded/iq1m`; do
       uv run roverd checksum /path/to/downloaded/iq1m/$i /path/to/checksums/output/$i;
   done
   

2. Compare the calculated checksums against the reference checksums:

   for i in `roverd list /path/to/downloaded/iq1m`; do
       uv run roverd checksum /path/to/original/checksums/$i /path/to/checksums/output/$i;
   done