Description

The Soft-Biometric in Surveillance (SoBiS) dataset was collected to provide a unified dataset to evaluate the following four computer vision tasks:

• person detection,
• person tracking,
• person re-identification and
• soft-biometric attribute recognition.

The dataset was recorded in a camera network at Fraunhofer IOSB and contains many challenges that computer vision algorithms and surveillance applications have to deal with.
Video data was recorded by seven cameras on three days and contains 31 volunteers which reappear multiple times. Scenes range from low complexity with only one person in the camera network to high complexity witch 15 or more persons in the recorded area. For a more detailed description of the dataset please see the corresponding publication below.

Contents

The dataset contains the following items:

• Video data for all seven cameras (no sound)
• Manual person annotations every 5 frames
• Consistent person IDs over the entire dataset
• Clothing-specific IDs to indicate when a person altered their visual appearance
• Soft-biometric attribute annotation for each person track
• Camera calibration information
• A set of person detections to start from

Impressions

Download

We are obligated to keep track of who obtains the dataset and ensure that it is only used for academic purposes. Thus, the dataset is only available by request.
To request a download link for the dataset please send an email to .
Note: Due to recent data protection guidelines, we are no longer distributing this dataset.

Citation

Please cite this dataset as: