SHORT-100

The SHORT-100 dataset of grocery products is an updated and practical dataset for studying object recognition and retrieval in the challenging scenarios of hand-held objects and mobile or wearable cameras, and with emphasis in the assistive case for blind and partially sighted users.

One of the motivations for introducing SHORT is the possibility of applying visual object recognition to assistive applications, supporting people with visual impairment. The ubiquity of camera-equipped smartphones, and their accessibility options, make them potentially well suited to this application.

SHORT-100
SHORT-100 products
Dataset details

100 categories of products found on a typical shopping list.

Training dataset – 36 high quality images taken at 3 elevations and 12 different angles in a controlled setup with a Nikon D7100 SLR.

Query dataset – 134,524 images crowdsourced with 30 different smarpthones. The queries are divided in sub-datasets of still images, video frames, acquired by sighted users or blindfolded.

Download
Publications

If you use this dataset, please kindly cite one of the following publications:

  • J. Rivera-Rubio, S. Idrees, I. Alexiou, L. Hadjilucas, and A. A. Bharath, “A dataset for hand-held object recognition,” in 2014 IEEE International Conference on Image Processing (ICIP), Paris, 2014, 4 pages.
    [BibTeX]
    @inproceedings{Rivera-RubioICIP,
    address = {Paris},
    author = {Rivera-Rubio, Jose and Idrees, Saad and Alexiou, Ioannis and Hadjilucas, Lucas and Bharath, Anil A.},
    booktitle = {2014 IEEE International Conference on Image Processing (ICIP)},
    pages = {4 pages},
    title = {{A dataset for hand-held object recognition}},
    year = {2014},
    keywords = {CV}
    }

  • J. Rivera-Rubio, S. Idrees, I. Alexiou, L. Hadjilucas, and A. A. Bharath, “Small Hand-held Object Recognition Test (SHORT),” in 2014 IEEE Winter Conference on Applications of Computer Vision (WACV), 2014, pp. 524-531. doi:10.1109/WACV.2014.6836057
    [BibTeX]
    @INPROCEEDINGS{6836057,
    author={Rivera-Rubio, J. and Idrees, S. and Alexiou, I. and Hadjilucas, L. and Bharath, A. A.},
    booktitle={2014 IEEE Winter Conference on Applications of Computer Vision (WACV)},
    title={Small Hand-held Object Recognition Test ({SHORT})},
    year={2014},
    month={3},
    pages={524-531},
    keywords={CV;cameras;object recognition;smart phones;SHORT;assistive system context;database datasets;grocery products;ground truth;high quality cameras;high quality catalogue images;high quality training images;high-street shopping;price comparisons;query datasets;small hand-held object recognition test;smartphone-captured test images;variable quality user-captured images;visual impairment;visual object recognition;wearable cameras;Cameras;Context;Databases;Object recognition;Training;Videos;Visualization},
    doi={10.1109/WACV.2014.6836057},
    }

Code and other Data

Jose Rivera has developed an easy to use MATLAB version of a typical object categorization pipeline that can be used to extract SIFT/DSIFT features, cluster the features, try out different encoding techniques and perform SVM classifications. The SHORT-100 dataset can be used straight away with this code.

The code is hosted in Bitbucket: https://bitbucket.org/josemrivera/bow_pipeline_toolkit

git clone https://josemrivera@bitbucket.org/josemrivera/bow_pipeline_toolkit.git

 

Coming soon: descriptor data

Poster: PDF

Acknowledgements

We thank our friend and expert photographer Alberto Cerveto for his invaluable help during the acquisition of the training images and advice when preparing the setup.

We also thank the many volunteers that helped collecting the test images, especially those from the  HIPR – Image Processing course that happily joined after the labs.