Datasets

  1. The RSM dataset  This is a visual dataset of indoor spaces to benchmark localisation/navigation methods.  It consists of around 10 GB of video data capturing more than 1.5 km of corridors and indoor spaces within the RSM building at Imperial College London. The data was collected in 6 corridors, each travelled five times in order to allow for repeatability studies. The database includes ground truth for every frame, measured as distance in centimetres from starting point. Authors:  J. Rivera-Rubio, I. Alexiou and A. A. Bharath; Originally posted: May-2014; See the Website or Browse.

  2. SHORT-100, Small Hand-held Object Recognition Test, a new dataset that aims to benchmark the performance of algorithms for recognising hand-held objects from either snapshots or videos acquired using hand-held or wearable cameras.  SHORT is designed to be focused on the assistive systems context, though it can provide useful information on more general aspects of recognition performance for hand-held objects. The present state of the dataset is comprised of a small set of high quality training images and a large set of nearly 135,000 smartphone-captured test images of 100 grocery products. Authors:  J. Rivera-Rubio, S. Idrees and A. A. Bharath; Originally posted: May-2014; See the Website or Browse.

  3. Multimodal Gait Dataset  This will soon be available; we are still sorting out the details of using a new data repository.  The dataset contains image sequences and MoCap (motion capture) data of people walking and running on a treadmill at different speeds (3-12km/h). Can be used for tests into gait recognition. MoCap can also be used for synthesizing images under the effects of different factors that can alter the visual appearance of a subject, whilst maintaining the motion captured from real human subjects. For example, using avatars (provided), one can generate sequences from different views, with different clothing, lighting, camera parameters etc. Such data can be used to train networks to be less dependent on confounding factors that can reduce discrimination performance in biometrics.  Data collected and curated by  C. Charalambous. Paper is available here.