TOPSPIN: daTaset Of sPinning objectS with neuromorPhic vIsioN

Dataset used in the paper "Event-based Structure-from-Orbit", for the purpose of sparse point cloud reconstruction of a spinning object.

The dataset consists of six objects rotating at three speeds, being viewed by a static event camera from four different perspectives. For each scene, one of the objects was placed on the turntable and rotated at a given speed, while the camera observed it from a given perspective. We exhaustively enumerated all combinations of objects, rotational speeds and perspectives to generate 72 scenes in total.

Scene names use the following encoding: {object}-{perspective}-{speed}

All scenes have the same data format:

scene/

    events.csv -- Event stream with the format {timestamp, x, y, polarity (0=off, 1=on)}

    calib.txt -- Camera calibration file, with the format "fx fy cx cy k1 k2 p1 p2 k3"

    meta.json -- Metadata file with camera frame dimensions

Note: all timestamps are in microseconds.

When using the data in an academic context, please cite the following paper.

@inproceedings{elms2024eventbased,
      title={Event-based Structure-from-Orbit}, 
      booktitle={IEEE Conf. on Computer Vision and Pattern Recpgnition},
      author={Ethan Elms and Yasir Latif and Tae Ha Park and Tat-Jun Chin},
      year={2024},
}