attentional_RBM.html

<h2 id="MVL">Attentional tracking with deep learning</h2>

  <img class="paper" title="CVIU 2013" src="./page_files/RBM_ICML.png"><br>
    <a class="paper" href="http://lorisbaz.github.io/papers/proceedings/Deniletal_NECO12.pdf"> Learning where to attend with deep architectures for image tracking</a><br>
    M. Denil, <strong>L. Bazzani</strong>, H. Larochelle, and N. de Freitas <br>
    Neural Computation, 2012 <br><br>

    <a class="paper" href="http://lorisbaz.github.io/papers/proceedings/Bazzanietal_ICML11.pdf"> Learning attentional policies for object tracking and recognition in video with deep networks</a><br>
    <strong>L. Bazzani</strong>, N. de Freitas, H. Larochelle, V. Murino, J-A Ting<br>
     The 30th International Conference on Machine Learning (ICML), 2011<br>


<h3>Details</h3>
<div class="paper">
In this papers, we propose a novel attentional model for simultaneous object tracking and recognition that is driven by gaze data. Motivated by theories of the human perceptual system, the model consists of two interacting pathways: ventral and dorsal. The ventral pathway models object appearance and classification using deep (factored)-restricted Boltzmann machines. At each point in time, the observations consist of retinal images, with decaying resolution toward the periphery of the gaze. The dorsal pathway models the location, orientation, scale and speed of the attended object. The posterior distribution of these states is estimated with particle filtering. Deeper in the dorsal pathway, we encounter an attentional mechanism that learns to control gazes so as to minimize tracking uncertainty. The approach is modular (with each module easily replaceable with more sophisticated algorithms), straightforward to implement, practically efficient, and works well in simple video sequences.
</div>


<div class="paper">
<center><h3><a href="https://github.com/lorisbaz/AttentionalRBM">Download Attentional Tracker</a></h3> (link to github)</center> <br>
<center><h4><a href="http://lorisbaz.github.io/datasets/ICML11_synth_dataset.7z">Download Synthetic dataset</a></h4></center> <br>
See the instructions in the README.md file.
</div>

<h3>Some cool videos</h3>
<div class="paper">
<center><iframe src="https://www.youtube.com/embed/aeHrBNAn7Yo" allowfullscreen="" frameborder="0" height="465" width="620"></iframe> <br><br>
<iframe src="https://www.youtube.com/embed/O3IUEmmhnmg" allowfullscreen="" frameborder="0" height="465" width="620"></iframe></center>
</div>