I/Q-1M: one million i/q frames

Info

We are currently working to prepare the dataset for public release and distribution. For the time being, please contact Tianshu Huang (tianshu2@andrew.cmu.edu) for access.

Overview

The I/Q-1M dataset consists of 1M radar-lidar-camera samples¹ over 29 hours across indoor, outdoor, and bike-mounted settings, each with a mobile observer:

• indoor: inside buildings at a slow to moderate walking pace, visiting multiple floors and areas within each.
• outdoor: neighborhoods ranging from single family detached to high density commercial zoning at a moderate to fast walking pace.
• bike: bike rides in different directions from a set starting point with a moderate biking pace.

Tip

See our paper for more details about the dataset. Make sure to download the arxiv version to see the attached (and linked) appendix!

Setting	Size	Length	Average Speed	Max Doppler	Max Range
indoor	310k	8.9h	1.0m/s	1.2m/s	11.2m
outdoor	372k	10.7h	1.4m/s	1.8m/s	22.4m
bike	333k	9.3h	5.4m/s	8.0m/s	22.4m

Index of Files

Tip

See the roverd documentation for details about the data format.

{sequence}
 ┣ 📂_camera
 ┃ ┣ 📜meta.json
 ┃ ┣ 📜pose.npz             # interpolated cartographer poses with camera timestamps
 ┃ ┣ 📜segment              # lzma-compressed semantic segmentation class maps
 ┃ ┣ 📜segment_i            # byte offsets
 ┃ ┗ 📜ts                   # camera timestamps (same as camera/ts)
 ┣ 📂_lidar
 ┃ ┗ 📜pose.npz             # cartographer poses with lidar timestamps
 ┣ 📂_radar
 ┃ ┗ 📜pose.npz             # cartographer poses with radar timestamps
 ┣ 📂_slam
 ┃ ┗ 📜trajectory.csv       # raw cartographer output
 ┣ 📂camera
 ┃ ┣ 📜meta.json
 ┃ ┗ 📜ts                   # camera timestamps (30Hz)
 ┣ 📂imu
 ┃ ┣ 📜acc                  # linear acceleration
 ┃ ┣ 📜avel                 # angular velocity
 ┃ ┣ 📜meta.json
 ┃ ┣ 📜rot                  # rotation
 ┃ ┗ 📜ts                   # IMU timestamps (100Hz)
 ┣ 📂lidar
 ┃ ┣ 📜lidar.json
 ┃ ┣ 📜meta.json
 ┃ ┣ 📜nir                  # lzma-compressed near-infrared image
 ┃ ┣ 📜nir_i                # byte offsets
 ┃ ┣ 📜rfl                  # lzma-compressed IR reflectance
 ┃ ┣ 📜rfl_i                # byte offsets
 ┃ ┣ 📜rng                  # lzma-compressed beam-time depth map
 ┃ ┣ 📜rng_i                # byte offsets
 ┃ ┗ 📜ts                   # lidar timestamps (10Hz)
 ┣ 📂radar
 ┃ ┣ 📜iq                   # raw complex time signal
 ┃ ┣ 📜meta.json
 ┃ ┣ 📜radar.json           # radar intrinsics
 ┃ ┣ 📜ts                   # radar timestamps (20Hz)
 ┃ ┗ 📜valid                # whether frames contain zero-filled dropped packets
 ┗ 📜config.yaml            # original data collection configuration

Semantic Segmentation Classes

0	1	2	3
flat	nature	sky	structure

4	5	6	7
ceiling	object	person	vehicle

For full details about the class definitions, see the class mapping and original ADE20k dataset.

---

1. The radar was collected at 20Hz, the Lidar at 10Hz, and the camera at 30Hz; as such, Lidar is limiting sensor to arrive at our 1M sample count. ↩