Dear All

All are invited for the same

What : Talk on Object Detection in Presence of Hard Examples
When : 11 March 2016 , 5:00 PM - 6:00 PM
Where : KD 101 (HRKD Building)
Who : Subhabrata Debnath

Object detection involves finding the location of the object of interest in
an image. This is done by learning a detector using training images
containing the location of the object as input. Thus the robustness of the
detector directly depends on the quality of training set. In our work we
try to achieve robust detection even in the presence of visually hard
images in the training data. In our setting, we are presented with labels
which indicate the presence or absence of an object in an image but not
their explicit locations. This is called a weakly supervised setting. We
aim to learn a detector which can classify a test image as well as find the
location of the bounding box containing the object in the image. This can
be done by modeling the location of the object as a latent parameter and
learning both the location and the classifier jointly during training. We
show how using a variation of Outliers Robust-SVM and Self paced learning
with latent variables can be used to obtain good results in this scenario.
We show our results on three classes of the Pascal Voc 2007 dataset and
present a comparison with existing methods.

*Bio :* Subhabrata Debnath is Co-Founder and Computer Vision Researcher @
VisageMap Inc. He previously worked in Information Systems Officer at
Indian Oil Corporation Limited(IOCL) . He did his  Mtech in CSE 2013 -2015
from  IIT Kanpur  and  Btech from Bengal Institute of Technology, Kolkata
in 2008 - 2012. Publication : Adapting Ransac SVM to detect outliers for
robust classification, BMVC 2015, [Subhabrata Debnath, Anjan Banerjee,
Vinay Namboodiri](

Yours Sincerely
Vivek Gupta