Ghanem, Convolution Neural Net Attributes, KAUST, CVPR, 2015
<p>Jun 27, 2015 12:34 PM</p>
<p>
Victor Escorcia seems to have written the paper (altho Castillo is listed )
KAUST King Abdullah Univ of Sci and Tech.</p>
<p>
from CVPR 2015 (conf was in Boston this year)</p>
<p>
One of the cornerstone principles of deep models is their abstraction capacity, i.e. their ability to<b> learn abstract concepts from ‘simpler’ ones</b>.</p>
<p>
Through extensive experiments, we characterize the nature of the relationship between abstract concepts (specifically objects in images) learned by popular and high performing convolutional networks (conv-nets) and established mid-level representations used in computer vision (specifically <b>semantic visual attributes</b>).</p>
<p>
We focus on attributes due to their impact on several applications, such as object description, retrieval and mining, and active (and zero-shot) learning. </p>
<p>
Among the findings we uncover, we show empirical evidence of
the existence of <b>Attribute Centric Nodes (ACNs) </b>within a conv-net, which is trained to recognize objects (not attributes) in images. </p>
<p>
These special conv-net nodes (1) <b>collectively encode information pertinent to visual attribute representation and discrimination, (2) are unevenly and sparsely distribution across all layers of the conv-net,</b> and (3) play an important role in conv-net based object recognition.
</p>
<p>
</p>
<p>
</p>