All I Have Seen (AIHS)
by Nebojsa Jojic, Alessandro Perina, Vittorio Murino,

Neural Information Processing Systems (NIPS), 2010

[ Alessandro's home page ] [ Nebojsa's home page ]


Abstract

In order to study the properties of total visual input in humans, for around two weeks Nebojsa wore a camera capturing, on average, an image per every 20 seconds of his waking hours. The resulting new dataset contains a mix of indoor and outdoor scenes as well as numerous foreground objects. Our first analysis goal is to create a visual summary of the subjectís two weeks of life using unsupervised algorithms that would automatically discover recurrent scenes, familiar faces or common actions. Direct application of existing algorithms, such as panoramic stitching (e.g.  Photosynth) or appearance-based clustering models (e.g. the epitome), is impractical due to either the large dataset size or the dramatic variation in the lighting conditions. We dubbed this type of data "All I have Seen" (AIHS, meant to be pronounced similar to "eyes"). While these types of datasets have been assembled before, it is our belief that with the proliferation of mobile devices and the availability of cloud computing, the time is now more appropriate than ever for research into this type of data acquisition, unsupervised techniques for data analysis and applications on top of them.

This is Nebojsa's personal data, made available for research purposes only. If you use it, please cite:

@incollection{NIPS2010_1298,

title = {Structural epitome: a way to summarize oneís visual experience},

author = {Nebojsa Jojic and Alessandro Perina and Vittorio Murino},

booktitle = {Advances in Neural Information Processing Systems 23},

editor = {J. Lafferty and C. K. I. Williams and J. Shawe-Taylor and R.S. Zemel and A. Culotta},

pages = {1027--1035},

year = {2010}

}

download


Labeled dataset

Alessandro manually labeled a subset containing 300 images and we used it to evaluate some of our first solutions (NIPS 2010). We are currently labeling a larger random subset (approx 4500 images ) with a variety of labels.

download


Full dataset

download statistics


Below, for each day we show 10 random images.

Click on an image to download the matfile.


Day 1 : 21st April 2009 ( 304 Images )
Day 2 : 22nd April 2009 ( 3617 Images )
Day 3 : 23rd April 2009 ( 1080 Images )
Day 4 : 24th April 2009 ( 2780 Images )
Day 5 : 27th April 2009 ( 1334 Images )
Day 6 : 28th April 2009 ( 4104 Images )
Day 7 : 29th April 2009 ( 2193 Images )
Day 8 : 30th April 2009 ( 2176 Images )
Day 9 : 1st May 2009 ( 1526 Images )
Day 10 : 4th May 2009 ( 3700 Images )
Day 11 : 5th May 2009 ( 3600 Images )
Day 12 : 7th May 2009 ( 2049 Images )
Day 13 : 8th May 2009 ( 2360 Images )
Day 14 : 10th May 2009 ( 3761 Images )
Day 15 : 12nd May 2009 ( 2177 Images )
Day 16 : 14th May 2009 ( 4392 Images )
Day 17 : 15th May 2009 ( 1098 Images )
Day 18 : 19th May 2009 ( 1815 Images )
Day 19 : 10th July 2009 ( 1546 Images )