DSEC-MOT: an Event-based MOT Dataset
- Citation Author(s):
-
Song Wang
(The University of Hong Kong)
Zhu Wang
(The University of Hong Kong)
Xiaojuan Qi
(The University of Hong Kong)
Can Li
(The University of Hong Kong)
Hayden Kwok-Hay So
(The University of Hong Kong)
- Submitted by:
- Song Wang
- Last updated:
- DOI:
- 10.21227/kg81-z613
- Data Format:
- Research Article Link:
- Links:
73 views
- Categories:
- Keywords:
Abstract
Introduction
Welcome to the new challenging event-based multi-object tracking dataset (DSEC-MOT) repository. Our goal is to provide a challenging and diverse event-based MOT dataset with various real-world scenarios to facilitate the objective and comphrehensive evaluation of event-based multi-object tracking algorithms. This dataset, built upon DSEC, contains a variety of traffic entities and complex scenarios, aiming to address the current lack of event-based MOT datasets.
Dataset Statistics
DSEC-MOT contains:
- 12 annotated sequences with a total of 37,080 boxes, 502 tracks, and 769.2-second duration
- 7 object classes including cars, pedestrians, bicycles, motorcycles, buses, trucks, and trains
- Challenging scenarios including crowded urban areas and rural scenes with various lighting conditions
- Severe occlusions, making the dataset suitable for testing trackers' occlusion handling capabilities
- Event resolution 640 x 480
Instructions:
Download
Annotation files (1.0 MB)
Annotation Format
The annotation data for the DSEC-MOT dataset is stored in txt files, with each annotation occupying a single line.
The format of the annotation is as follows:
`[timestamp, track ID, box left, box top, box width, box height, class ID]`
- `timestamp` represents the time at which the specific object is annotated
- `track ID` is assigned for each unique object being tracked in the event sequence
- `box left` and `box top` denote the coordinates of the top-left corner of the bounding box
- `box width` and `box height` specify the dimensions of the bounding box surrounding the object
- `class ID` indicates the class of the object being tracked
| Class ID | Label |
|-------------|-----------------|
| 0 | Cars |
| 1 | Pedestrians |
| 2 | Bicycles |
| 3 | Motorcycles |
| 4 | Buses |
| 5 | Trucks |
| 6 | Trains |
Example annotation:
`[54828357559, 10, 178, 222, 100, 70, 0]`
This example annotation represents a car appearing at the timestamp 54828357559 microsecond, with a unique track ID of 10. The bounding box for this object has its top-left corner at coordinates (178, 222) and dimensions of width 100 and height 70.
We hope our DSEC-MOT dataset contributes to advancing research in event-based multi-object tracking. If you have any questions or issues, feel free to reach out to us or open an issue in this GitHub repository. Enjoy tracking!
