PHASE IV

Presenter Name	Nikunj Patel and Tushar Kataria, Final year MTech, DSP lab
Cluster	EE1
Email-id:
Topic:	Image hallucination: Be there at all times!
Venue:	GG 301
Session Chair:
Date:	17/03/2017 at 4 PM
Pre-Requisite:	None
Abstract:	Image hallucination is an exciting field in which one can simulate different times of the day. Specifically, say, if one captures an image of Dubai skyline during day, it is possible to get the same image showing Dubai skyline during night. Image Hallucination has many applications in areas such as computational photography and image fusion. This talk will explore an image hallucination technique based on the template (patch) matching from the database of time lapse images and learned locally affine model. The color transfer model, by hallucinating a variety of natural images at different times of the day, will be presented.
Presentation Slides:

Presenter Name	Saurabh Kumar Final year M.Tech, Vision and Image Processing Lab
Cluster	EE1
Email-id:	saurabhkm@ee.iitb.ac.in
Topic:	Deep diving with Caffe
Venue:	GG 301
Session Chair:	M. Fasil
Date:	25/01/2017 at 4.15 PM to 5.15 PM
Pre-Requisite:	None
Abstract:	Ever wondered what goes inside the current driver-less cars, speech recognition systems or face recognition systems? One will have to go really 'deep' to understand this. All such systems include Deep networks. This talk will go over a few basics of Artificial Neural Networks and build upon them to introduce Deep Network. The talk will have a demonstration on how one can build powerful deep learning architectures with the Caffe, the popular deep learning framework. The main take away from this talk will essentially be on how one can train deep net using Caffe to recognize images and also adapt it to solve different research problems. The session will conclude with discussion on how this setup has been used to recognize 3D objects from point cloud data.
Presentation Slides:

Presenter Name	Ashwin Lele & Saurabh Pinjani
Cluster	EE1
Email-id:
Topic:	The search for similarity- A musical investigation
Venue:	GG 302
Session Chair:	Kaustuv Kanti Ganguli
Date:	15/02/2017 at 2.30 PM
Pre-Requisite:	None
Abstract:	If you like a particular tune fragment from a song and want to know where it lies in the entire song, how would one go about this problem? Or can we retrieve the instances of a particular phrase from a large dataset just by humming it? How can we encode the melody of a song compactly without losing relevant information? We address some of these problems in the context of Hindustani classical vocal music. We analyze the design considerations at each stage: pitch curve fitting, normalization, discrete symbol representation and string matching cost functions. A combination of domain knowledge and data-driven optimization on a database of raga music is exploited to design the melodic representation of a raga phrase that achieves high retrieval accuracy at a low computational cost.
Presentation Slides:

Presenter Name	Anil Kamma
Cluster	EE1
Email-id:
Topic:	Multi-Mode Resonator based Ultra Wide Band and Multiband Filters.
Venue:	GG 303
Session Chair:	Hemant Kumar
Date:	17/02/2017 at 3.30 PM
Pre-Requisite:	None
Abstract:	With the rapid growth of communication electronics, efficient spectrum utilization has become quintessential. This demands highly precise Ultra Wide Band (UWB) and multi band filters. A novel design approach for such filters based on multi-mode resonators will be illustrated in this talk, with emphasis on a notch filter designed and tested by the speaker.
Presentation Slides:

Presenter Name	Nikunj Patel and Tushar Kataria, Final year MTech, DSP lab
Cluster	EE1
Email-id:
Topic:	Image hallucination: Be there at all times!
Venue:	GG 301
Session Chair:
Date:	17/03/2017 at 4 PM
Pre-Requisite:	None
Abstract:	Image hallucination is an exciting field in which one can simulate different times of the day. Specifically, say, if one captures an image of Dubai skyline during day, it is possible to get the same image showing Dubai skyline during night. Image Hallucination has many applications in areas such as computational photography and image fusion. This talk will explore an image hallucination technique based on the template (patch) matching from the database of time lapse images and learned locally affine model. The color transfer model, by hallucinating a variety of natural images at different times of the day, will be presented.
Presentation Slides: