UWCR Dataset - RADAR MINI

UWCR RaDAR mini dataset is created by ECE at the University of Washington and CMMB Vision. Please acknowledge the dataset source and provide a copy of the paper/report if you use our dataset.

Download

To get access to our dataset, please first fill in a permission request form using the following link: https://forms.gle/DsYnn84BtVBJi2bz9. We will send the download link through email after approval.

Summary

There are 9 RaDAR sequences in this dataset. Each sequence has a framerate of 30 FPS, 255 chirps per frame, and length around 900 frames

The dataset stucture is shown below.

━ UWCR_RADAR_MINI
    ┗ data
        ┗ 2019_04_09
            ┗ 2019_04_09_cms1000
                ┣ radar
                ┗ ramap_labels.csv
            ┗ ... (sequence names)
            ┗ 2019_04_09_pss1003
    ┗ scripts

For each data sequence, RaDAR data are in radar folder with *.bin format. Metadata annotations are stored in *.csv format.

Data Collection

Our dataset is collected using our self-build sensor platform shown below.

Usages

RaDAR Data Preprocessing

Written in MATLAB format in parse_radar.m.

Metadata Parser

Metadata parser script is in Python format in scripts folder named radar_anno_loader.py. Sequence path can be set inside the script. The result return from function read_ra_labels_csv(seq_path) can be described as follows:

Return a list of object annotation with length of frame number.
Each element of the list is [range_idx, angle_idx, class_id].
The mapping function for range/angle index to real value is written in mappings.py.

tomfisher / UWCR_Dataset_RADAR_mini